The Impact of Illegal Insider Trading in Dealer and Specialist Markets: Evidence from a Natural Experiment
Raymond P.H. Fishe
and
Michel A. Robe
School of Business Administration University of Miami P.O. Box 248126 Coral Gables, FL 33124
Kogod School of Business American University 4400 Massachusetts Avenue, NW Washington, DC 20016
[email protected] (305) 284-4397
[email protected] (202) 885-1880
January 2002
The authors thank the officials at the Securities and Exchange Commission and the U.S. Attorney’s Office in New York for assistance with the study. The authors thank Jim Angel, Henk Berkman, Graeme Camp, Jeff Harris, Kris Jacobs, Tim McCormick, Ron Melicher, Albert Minguet, David Reeb, and seminar participants at the NASD, the University of Auckland, McGill University, and the 2001 Meetings of the European Finance Association (Barcelona) and of the Financial Management Association (Toronto) for helpful comments. They are grateful to Tim McCormick for help in obtaining aggregate depth data for Nasdaq-listed stocks. This work began while Pat Fishe was a Visiting Academic Scholar at the Securities and Exchange Commission. The Securities and Exchange Commission, as a matter of policy, disclaims responsibility for any private publication or statement by any of its employees. The views expressed herein are those of the authors and do not necessarily reflect the views of the Commission or the authors’ colleagues upon the staff of the Commission. Michel Robe would like to acknowledge the support received as a Kogod Endowed Research Fellow. Xinxin Wang provided valuable research assistance. The authors are responsible for all errors and omissions.
1
The Impact of Illegal Insider Trading in Dealer and Specialist Markets: Evidence from a Natural Experiment
Abstract This paper provides direct evidence on market makers’ reaction to unambiguously informed trading in specialist versus dealer markets. Using the trades of stockbrokers who had advance copies of a stock analysis column in Business Week magazine, we document that increases in price and volume occur after informed trades and before public release of the information, especially for Nasdaq stocks. Both quoted and effective bid-ask spreads are unaffected by informed trades. Instead, market makers adjust the depth at the ask quotes as the information leads to more buys. Quoted depth falls once insider trading begins and then rebounds after it ends, generally to above its initial level. Ask depth falls relatively more on the NYSE and AMEX than on the Nasdaq, which suggests that specialist markets detect informed trading more readily. None of these pre-release changes are observed in a control sample of stocks that were mentioned in the column but not traded by these stockbrokers. Overall, our results show that illegal insider trading has a negative impact on market liquidity and that market makers use depth as the tool to manage asymmetric information risk during unexpected insider trading episodes. JEL-Classification: G12, G14, K22, D82 Keywords:
Insider Trading, Asymmetric Information, Depth, Liquidity, Specialist vs. Dealer Market, Business Week
2
I.
Introduction Insider trading in financial markets has been the focus of extensive study for many years.
Central to the legal and economic debates surrounding this issue are key empirical questions. What is the impact of insider trading on stock prices and trading volume? Can market makers detect the presence of privately informed traders? If so, how does market liquidity—not only spreads but also depth—react to the onset of insider trading? There exist, however, very few empirical studies of unambiguously informed trading. Most studies rely on the position of a trader (e.g., company official or board member) to infer insider information. This paper contributes to the debates by analyzing a recent insider trading case involving information on 116 stocks obtained by five stockbrokers. Their “inside” knowledge was gained by acquiring copies of Business Week’s “Inside Wall Street” (IWS) column in advance of public release. These stockbrokers were not privy to any news emanating directly from the corporations whose shares they traded. In effect, they were outsiders trading on second-hand, non-public, short-lived information about specific firms. Nevertheless, trades based on advance knowledge of the column yielded significant abnormal returns. Because the “insiders" (the stockbrokers and their associates) traded only a third of the stocks mentioned in the column over an eight-month period, this episode offers a unique natural experiment to study the impact of genuinely informed trading. Furthermore, because the stocks involved were listed on the Nasdaq, NYSE and AMEX, the data provide the first opportunity to contrast how illegal insider trading by the same parties affects specialist and dealer markets. We find strong evidence that illegal insider trading has a negative impact on liquidity and that market makers adjust depth—not bid-ask spreads—to manage the risk presented by privately informed traders.1 These results hold for both specialist and dealer markets. The difference between market structures arises in the magnitude of the depth response. Specialist markets reduce depth much more than dealer markets in response to insider trading, which suggests that 1
Throughout the paper, we use the term “market makers” to denote all liquidity providers, including specialists, dealers and limit-order traders.
specialist markets are better equipped to detect such trading (cf., Heidle and Huang, 2002). The events analyzed here were publicly revealed in January 1999, when the SEC charged five stockbrokers with insider trading on misappropriated nonpublic information from Business Week magazine.2 The SEC alleged that one of the brokers, Larry Smath, paid foremen of the local Business Week distributor, Hudson News Co., to fax him advance copies of Gene Marcial’s IWS column. He obtained this information in the early afternoon on Thursdays, before the public release of portions of the magazine over news wire (typically at 5:15 PM the same day) and electronic distribution on America Online (at 7 PM). Smath was able to forward it to the other brokers and they were able to enter trades before the markets had closed. The SEC charged that this scheme involved trades in at least 39 different stocks between June 1995 and January 1996, and apparently ended only because officials at Business Week noticed unusual trading in some of the recommended stocks before the magazine’s release.3 In all, the defendants, members of their families and some of their clients bought $7.73 million worth of securities mentioned in the IWS column, accounting for about 5 percent of total Thursday trading in the affected stocks.4 This case is of general interest because it offers a close-up view of market making during numerous episodes of unambiguously informed trading. For every stock traded by the insiders, as well as for most of the stocks mentioned in the IWS column that the insiders did not trade, we have data about all transactions (trade time, volume and price, execution market) and quotes (bid and ask prices, quoted bid and ask depths) for three days around the insider trading day, which was always a Thursday. Court records from the civil and criminal cases brought against the brokers identify their trades within the transaction stream. By aggregating and analyzing the trade and quote data in 15-minute intervals, we obtain a detailed picture of investors’ and market makers’ behavior during, and immediately following, periods of insider trading activity.
2
See e.g. “Group of Brokers is Facing Charges of Insider Trading,” The New York Times, January 28, 1999, p. C-21. See “Is Someone Sneaking a Peek at Business Week? Early Trading of a Few Inside Wall Street Stocks Raises a Red Flag,” by Chris Welles, Business Week, February 5, 1996. 4 Smath and two other brokers pled guilty to one felony count each. Another broker, Joseph Falcone, was convicted of insider trading on November 9, 1999, following a 1 1/2-week trial. The fifth broker cited in the SEC complaint was never criminally charged; neither were the brokers' associates. 3
-2-
A key question examined in this study is how informed trades affect market liquidity. A basic tenet of market microstructure theory is that liquidity partially reflects the information asymmetry created by informed traders (Madhavan, 2000). Most microstructure models focus on bid-ask spreads as the tool to react to informed trading (e.g., Glosten and Milgrom, 1985; Glosten, 1989; Easley and O’Hara, 1992). Only recently have models explored whether market makers may change quoted depth as well as spreads in response to perceived increases in insider trading (Kavajecz, 1998; Dupont, 2000). Recent empirical work indicates that both spreads and depth are affected by expected information events.5 A natural question, however, is whether spreads or depth also react to unexpected changes in informed trading. To date, the sole evidence comes from case studies of two NYSE-traded stocks that were targeted by corporate insiders in the early 1980’s (Cornell and Sirri, 1992; Chakravarty and McConnell, 1997). Those authors find that liquidity, if anything, improves during insider trading episodes. Our findings are the first broader-sample evidence that genuinely informed trading has a negative impact on market liquidity. Further, our results indicate that market makers do not (or perhaps cannot) increase spreads in response to informed trading but do have the wherewithal to decrease depth. This result, which holds for Exchangelisted and (to a lesser extent) Nasdaq stocks, provides partial support for recent theoretical results (Dupont, 2000) that, in a specialist market, depth should react proportionally more than spreads to changes in informed trading.6 Specifically, we document that neither quoted nor effective spreads are affected by the arrival of informed trades, and establish this result for both specialist and dealer markets. Instead, we find that market makers limit their exposure to informed traders by reducing quoted depth. The data show that depth at the asked quote decreases during intervals of insider buying activity,
5
Liquidity falls in anticipation of and immediately following earnings announcements (e.g., Lee, Mucklow and Ready, 1993; Kavajecz, 1999), dividend announcements (Koski and Michaely, 2000) and takeover announcements (Foster and Vishwanathan, 1994; Jennings, 1994). See Kim and Verrecchia (1994) and Krinsky and Lee (1996) for discussions of earlier empirical studies analyzing spread behavior around such information events. 6 Kavajecz (1998) also explicitly models quantities and prices as separate choice variables. He forecasts that depth should fall and spreads should widen around an increase in the amount of adverse selection.
-3-
with Nasdaq depth declining less than Exchange-listed depth. Once insider trading ends, depth rebounds. Relative to the average quoted depth on the previous day, we find that ask depth is 38 percent lower for NYSE and AMEX stocks during insider trading intervals. In sharp contrast, after controlling for lower Nasdaq depth, the quoted ask depth for Nasdaq stocks falls by only 3 percent during insider intervals.7 The results are even stronger when we exclude nine traded stocks featured in non-Business Week news stories on the day before, or the morning of, the insider trading day. After removing those stocks, we find larger ask depth reductions, with a similar gap between the insider-related depth decreases for Exchange-listed stocks (-59 percent) versus their Nasdaq-listed counterparts (-19 percent). A salient feature of the present study is that it does not involve corporate insiders trading vast numbers of shares based on internal information. Instead, the “insiders” in the case were singling out firms after obtaining advance copies of a magazine column and buying a relatively small number of shares of each company selected on the basis of that short-lived information. This raises the questions of whether and how the private information gets impounded into prices. The private information involved had a very short useful life, so insiders were pressed into action in a relatively short trading window. We find that Thursday volume is not unusual up to the time of the first insider trade. During intervals when the brokers traded, however, there are significant increases in the number of trades, and there are further volume increases after the brokers are done trading. The insider-day volume increase is large (almost two-thirds of the previous day’s total volume), but the brokers’ trades only account for a small part of the increase. Court records imply that the IWS information was shared beyond the group of defendants charged by the SEC, but trades by the brokers’ associates only explain a fraction of the additional trading. Altogether, the trades of all the individuals identified by the SEC as possibly privy to some IWS information make up no more than 9.2 percent of the volume increase for insider-traded stocks. There is no reason to believe that, although it did not prosecute all of them,
7
For Nasdaq stocks, ask (bid) depth quotes are aggregated across all market makers quoting the best ask (bid) price, so that our Nasdaq depth figures are comparable to their counterparts for Exchange-listed stocks.
-4-
the SEC did not identify all the insiders in the case. Consistent with this conjecture, we find none of the effects discussed above for stocks that were mentioned in the IWS column but that the insiders did not trade. As such, a large part of the insider-day volume increase must be due to non-insider trades. We argue that this increased buying reflects noise trading by either “falsely informed” or mimicking and momentum traders. Overall, the buy-side activity is higher both during and after insider trading intervals, and prices rise markedly across these intervals. Consistent with the mimicking or momentum view, however, prices do not increase enough so that all of the information in the Business Week column is reflected in the Thursday closing price because abnormal returns are also observed on Friday. Lastly, a condition for trades based on secret knowledge of the IWS column to yield abnormal returns, was that stocks be promptly resold. We observe that the stockbrokers were slow to adjust their exit strategies and close their positions on the next day. They did learn this rule eventually, as their holding period consistently decreased during the period. The remainder of the paper proceeds as follows. Section 2 discusses related research. Section 3 describes the data and offers graphical evidence on the impact of insider trading. Section 4 analyzes the abnormal returns to insider-traded stocks mentioned in Business Week. Section 5 proceeds with the statistical analysis of trades, spreads and depth. Section 6 concludes.
II.
Relation to the Literature This paper is related to several strands of research. Foremost, it is part of a growing
empirical literature on the impact of insider trading in financial markets. Extant papers that study the impact of illegal insider trading focus on the trades of, overwhelmingly, corporate insiders or their tippees (Meulbroek, 1992; Cornell and Sirri, 1992; Chakravarty and McConnell, 1997, 1999). A key feature of the present paper is that the trades investigated were carried out by individuals who were outsiders, in that they did not have direct access to company records, plans, or confidential information: instead, they obtained their private information from a second-hand
-5-
source, Business Week. Those individuals traded only a third of the stocks mentioned in the IWS columns—with the other two thirds forming a unique, ideal control sample. Our study is likewise the first to contrast the impact of illegal insider trading in specialist and dealer markets. The pioneering study of Meulbroek (1992) uses SEC case files on illegal insider trading during the 1980-1989 period to determine if stock prices reacted to informed trading. Those files provide information on securities traded, volume and date of trades for 320 defendants and 218 different companies. Her final sample comprises 183 different cases of insider trading. She finds that the average cumulative abnormal return per insider trading episode is large (6.85 percent) and amounts to 47.6 percent of the abnormal return on the day the inside information becomes public.8 She also documents that the median insider’s trading represents only 11.3 percent of the affected stock’s total trading volume. Meulbroek makes a case, however, that insiders’ trades account for most of the extra trading volume on insider days, and hypothesizes that insider tradespecific characteristics—rather than volume per se—bring about the impounding of the inside information into security prices. Using intra-day data, we are able to document that trades are indeed different during insider purchasing intervals—they are much more numerous, yet smaller in size, than at other times and are overwhelmingly buyer-initiated. In contrast to Meulbroek’s findings, we find that insiders’ trades do not account for the major fraction of the trading volume increase on insider days. Overall, our evidence suggests that the latter is in large part due to a concomitant increase in noise trading by “falsely informed” or mimicking traders. While Meulbroek (1992) and the present paper deal with a cross section of insider trading cases, Cornell and Sirri (1992) and Chakravarty and McConnell (1997, 1999) analyze illegal trading by corporate insiders during two takeover attempts. Those case studies extend
8
Similarly, we find that the Thursday (i.e., insider-day) price increase for stocks traded by our five brokers is only a fraction of the total price increase following the release of the IWS column on Friday. Still, it is difficult to compare this result directly with Meulbroek (1992). On the one hand, it may be that the additional trading and related price increase on Thursday led naive IWS readers to believe that the column was all the more relevant, which should bring about an even bigger jump on Friday if readers traded on that basis. Indeed, we find a bigger overnight price jump for stocks traded by the brokers than for comparable IWS stocks that they did not trade. On the other hand, it may simply be that smaller firms, which make up a majority of the stocks traded by the brokers, routinely experience larger Friday IWS “bounces” than other firms in our control sample.
-6-
Meulbroek's work by analyzing the impact of insider trading not only on daily or hourly share prices, but also on bid-ask spreads and market liquidity. Specifically, Cornell and Sirri (1992) analyze 124 trades made by Paul Thayer, a director of Anheuser-Busch, and his accomplices during that company’s 1982 acquisition of CampbellTaggart. In all, 38 insiders bought a total of 265,600 shares over 23 trading days, which is equivalent to more than the monthly trading volume in six of the previous seven months and amounts to an “extraordinary” 29 percent of the target’s trading volume during that period. Unlike Meulbroek (1992), but consistent with our evidence, Cornell and Sirri document a large increase in non-insider, falsely informed trading over the same period. Nevertheless, they find that the abnormal returns on Campbell-Taggart stock occurred only on insider trading days. Using intra-day data, we show that the statistically significant positive returns on insider days actually arise after insiders finish trading. Cornell and Sirri’s most striking proposition is that insiders obtain superior execution for their trades, with the estimated bid-ask spread seemingly unchanged by insider trading. Further, they argue that liquidity improved while insiders were active, with liquidity measured as the cost of adding an additional share to an order rather than as changes in depth. We find support for the finding that spreads are not significantly affected by the onset of insider trading, and show that it generalizes to dealer markets. At the same time, however, we find little evidence of improved liquidity during insider trading episodes. To the contrary, we find that quoted depth falls sharply, with the strongest decrease affecting NYSE-listed stocks. The best-known insider case ever is arguably Ivan Boesky’s illegal purchase of 1,731,200 Carnation shares based on advance knowledge of Nestlé's 1984 acquisition of that firm. Chakravarty and McConnell (1997, 1999) analyze Boesky’s 366 trades, which were distributed over 24 days during 11 weeks. He accumulated almost 5 percent of Carnation’s outstanding shares. These authors document that Boesky’s trades made up half of the incremental volume, and that significant price increases took place both during and following hours when Boesky traded. Consistent with Cornell and Sirri (1992), they find that bid-ask spreads were hardly -7-
affected by Boesky’s transactions, even in the hours when he traded the most often or entered the largest trades. They also report that depth was unchanged or even improved during hours when Boesky bought shares, with quoted depth changes during “Boesky hours” greater on the bid side than on the ask side. They wonder, however, if “[those] results can or should be generalized to a larger population or to a different time period.” More generally, as pointed out in Chakravarty & McConnell (1997), a drawback of case studies is the uncertainty as to whether the results can be generalized. A key contribution of our paper is to demonstrate that, while many of the extant results can be reproduced in the context of a cross-section of insider trading episodes, some important results are not general in nature. In particular, we show that unambiguously informed trading has a negative impact on market liquidity, and that the magnitude of this impact depends on the type of financial market (specialist or dealer) where the trades are carried out. Our finding that the effects of illegal informed trading on market quality depends on market structure presents a counterpoint to Garfinkel and Nimalendran (2001). Those authors find that relative effective spreads are wider on days when insiders (legally or not) carry out trades than on non-insider days, and that the spread increase is stronger for NYSE stocks than for their Nasdaq counterparts. Such evidence is consistent with informed traders’ being more easily recognized by NYSE specialists than by Nasdaq dealers. In contrast, we find no consistent change in quoted or effective bid-ask spreads during (illegal) insider trading episodes, regardless of the market of execution. At the same time, we do find that such trades decrease quoted ask depth, and that these depth declines are also much stronger in specialist than dealer markets. One of our key objectives is to contrast the impact of (private) information in dealer and specialist markets and, in particular, the extent to which specialists and dealers detect the presence of informed traders. Hence, the present paper is also related to Corwin and Lipson (2000), Christie, Corwin and Harris (2002) and Heidle and Huang (2002). For the NYSE, Corwin and Lipson (2000) find that trading halts are sufficient to resolve price uncertainty because there are no systematic changes in price after the market reopens. In contrast, Christie, -8-
Corwin and Harris (2002) find that such halts do not resolve price uncertainty in a sample of Nasdaq stocks. In particular, spreads more than double following Nasdaq halts and only decrease after 20 to 30 minutes following the resumption of trading. Based on these findings, they argue that Nasdaq dealers, faced with incomplete knowledge of aggregate order flow, may be at a disadvantage to better-informed investors following halts. That conclusion is consistent with evidence that specialist markets appear better equipped to detect insider trades (Heidle and Huang, 2002). Our results provide further support for that interpretation, based on evidence from actual insider trades.9 Finally, to the extent that we focus on illegal trades based on positive news from advance copies of print media, our paper is also related to a sizable literature on the stock market impact of financial columns. The columns that have attracted academic attention are the Wall Street Journal’s “Heard on the Street” (e.g., Lloyd-Davis and Canes, 1979; Liu, Smith and Syed, 1990; and Beneish, 1991) and “Dartboard” columns (e.g., Barber and Loeffler, 1993; Greene and Smart, 1999; and Liang, 1999), as well as—the focus of the present paper—Business Week‘s “Inside Wall Street” column (e.g., Sant and Zaman, 1996). All studies of financial columns find significant positive excess returns when good news is reported.10 For favorable mentions in “Inside Wall Street,” average abnormal returns on the publication day ranged from 1.2 to 1.9 percent during the 1980’s. Using more recent data, we find abnormal returns more than twice that size, both six months before and during the insider-trading period. Sant and Zaman (1996), however, show that a favorable mention yields significant abnormal returns only for stocks that are followed by fewer than 21 analysts and that, for such stocks, the magnitude of the returns increases as the analyst following decreases.11 A key 9
Many other studies document differences in trading conditions between dealer and specialist markets. Most concentrate on differences in trading costs or items directly related to such costs. Examples include Huang and Stoll (1996), Barclay (1997), Bessembinder (1997, 1999), Bessembinder and Kaufman (1997a,b), Clyde, Schultz and Zaman (1997), LaPlante and Muscarella (1997), Barclay et al. (1999), Stoll (2000), Weston (2000), Chung, VanNess and VanNess (2001) and references cited in those papers. 10 The U.S. evidence presented in those papers is consistent with that from other countries. See, e.g., Wijmenga (1990) in the case of the Netherlands. 11 It is unlikely that the stockbrokers knew about Sant and Zaman’s research, as it was not published until 1996. However, they may have known of an earlier, well-publicized case of insider trading involving the same IWS column. In 1988, several security breaches occurred at Business Week. A number of people obtained advance copies
-9-
question, tackled by Sant and Zaman for the IWS column and by Greene and Smart (1999) and Liang (1999) for “Dartboard” picks, is whether that positive impact is long-lived. The answer is negative. In particular, Sant and Zaman (1996) find that the initial IWS announcement effect is negated after 26 trading days and that, within six months of a positive recommendation, these same stocks earn large negative abnormal returns.12 For stocks with the smallest analyst following (0 or 1-to-5 analysts), the loss exceeds 15 percent. Overall, making illicit gains based on advance access to the IWS column requires trading in stocks that have little analyst following and closing positions quickly.
III.
Data According to the criminal case filed by the U.S. Attorney and to the civil case filed by the
SEC, the scheme to obtain advance copies of Business Week’s IWS column started in June 1995 and ended with the February 5, 1996 issue.13 A total of 116 firms were mentioned in the column during that eight-month period. Court records provide information on securities traded by the five brokers and their associates; the date, volume and cost of each insider trade; and, for the brokers, the time of each trade and the profits earned from each transaction. Of the 116 firms, the stockbrokers did not trade in 76 companies, leaving 40 traded firms. We remove 10 companies from the traded sample: nine that were traded by a broker’s customer and but not by the brokers, which are missing time stamps, and one that had only stock options traded by one broker.14 The of the magazine from printing plants owned by R.R. Donnelley & Sons, and information was also leaked from within the company. Eleven individuals were convicted or settled charges of insider trading, including three stockbrokers and S.G. Ruderman, Business Week’s radio broadcaster, who went to prison for his participation. 12 Consistent with that result, Greene and Smart (1999) find that statistically and economically significant negative abnormal returns in the 29 days following publication of the “Dartboard” column erase all of the positive initial effect, leaving no evidence of persistent abnormal returns. Liang (1999) also finds strong mean reversion during the 15 days following publication, sufficient to erase the initial announcement effect for “Dartboard pro picks.” 13 See United States v. Joseph Falcone, 99 Cr. 332 (TCP) and SEC v. Smath et. al., 99 CV 523 (TCP). 14 The court records contain no information about the timing of the trades made by the brokers’ customers (who were never charged) or about the profits they made. Most of the customers’ IWS-based transactions were small. The main exception is a set of trades made by a named customer of two of the brokers. In addition to trading 21 of 30 stocks also purchased by the brokers, this customer traded nine IWS stocks that the brokers did not trade. Five of those additional trades were relatively small (ranging from 1,000 to 4,000 shares), and the other four trades (up to 8,000 shares) involved large companies mentioned in the IWS column: Conrail, MCI, American Express and Olin Corp. These nine stocks are removed from the sample.
- 10 -
focus is on the remaining 30 stocks that were traded by the five brokers, and for which there is complete data.
A.
Characteristics of the Traded Companies Some key characteristics of the firms traded by the brokers are summarized in Table 1.
The table also contrasts those firms with companies that were mentioned in the IWS columns but not traded by the stockbrokers. Data on the rates of return on assets and on equity, sales level and growth rate, assets, and growth rate of net income are from the 1994, 1995 and 1996 Compustat tapes.15 Information about company listing and the sentiment of the column (“Buy”, “Neutral” or “Sell”) is also included. Finally, the Dow Jones News Retrieval service (searching “all publications”) is used to determine whether featured firms were mentioned in other news articles on the Wednesday or Thursday preceding the public release of the IWS column. Table 1 Table 1 shows that the IWS column offered a favorable sentiment for the vast majority of stocks mentioned. In addition, most of these stocks were not mentioned in another news source. Thus, the Business Week column provides unexpected publicity for most of these companies. In the empirical analysis, we distinguish between companies with and those without other news to avoid the confounding effects that other news may cause. In addition, 45 percent of the traded firms were listed on the NYSE or AMEX, compared to 55 percent on the Nasdaq. Nearly the reverse of these listing proportions held for the non-traded firms. The Compustat data show that the companies traded by the stockbrokers were smaller than those not traded. The firms selected by the stockbrokers were also less profitable, although there is little difference in the growth rate of sales between traded and non-traded companies. The most striking differences are found in the size of these companies. The average sales of traded firms were less than one-half, and their average asset size was about one-fourth, of those 15
No Compustat data could be found for nine traded companies traded and 16 non-traded companies.
- 11 -
of their non-traded counterparts (median differences were somewhat smaller). The stockbrokers most likely anticipated that mention in the IWS column would have the largest impact on these smaller companies.
B.
Transaction and Quote Data For all 116 stocks mentioned in the IWS column, we collect transaction and quote data
from the Securities Industry Automation Corporation (SIAC) for three days beginning on Wednesday (the day before the IWS column was leaked), Thursday (the day of the leak), and Friday (the first day that the general public could trade on the information). The transaction data include time, volume and price, as well as the market on which each transaction was executed. The quote data include the bid and asked prices and depth, that is, the number of round lots for which each quoted price is “good.” The depth data for Nasdaq stocks is for all market makers quoting the best bid or ask price, which makes it comparable to exchange-listed depth figures. Trade direction is determined, for every transaction during each three-day period, with the Lee and Ready (1991) algorithm. For the regression analysis and figures, the data are summarized into 15-minute intervals, which smoothes the data and reduces the effect of larger trades and asynchronous trading on the results. The brokers’ trades are manually matched in the SIAC transactions stream. For most of the stocks traded, the information contained in court records (time, quantity, total cost) unambiguously identifies the brokers’ trades. Because some brokers’ orders were broken into smaller trades, the information from the court records does not always uniquely identify the trades. To address this problem, we identify all possible trade sequences that match the brokers’ trades around the time stamp and analyze the data in 15-minute intervals. It is rare for any trade sequence to cross between two 15-minute intervals. Even so, the statistical analyses are conducted across all sequences of insider trading intervals. The conclusions reported are robust across these alternative choices. Hence, in the remainder of the paper, we will only report results for regressions on the most likely candidate sequence. - 12 -
Table 2 Table 2 presents descriptive statistics of the SIAC data. The transaction information is reported in three panels. Panel (a) shows summary information for all 30 stocks traded by stockbrokers, Panel (b) shows information for 21 traded stocks that had no other news on either Wednesday or Thursday, and Panel (c) provides information on 44 non-traded stocks without any other news. Panels (a) and (b) show fairly similar statistics for most variables. Specifically, the average traded stock price is a little less than $20 with a quoted spread of about $1/4. Effective spreads range from $.12 to $.16 for these stocks. Across all three days, there are on average about 12 trades per 15-minute interval for traded stocks, with an average trade size of 1,550 to 1,750 shares. The average number of trades increases substantially from Wednesday (8.3 or 6.7) to Friday (17.1 or 15.1), but the average trade size shows a downward trend. This result is consistent with publicity created by mention in Business Week and with the findings of Sant and Zaman (1996) about the volume impact of the IWS column. The volume impact here is found in more, not larger trades, which is evidence that smaller investors are reacting to the IWS news. Changes in average depth across these days for traded stocks are shown in Panels (a) and (b) as well. In Panel (a), average ask depth is 8,600 shares on Wednesday, 8,000 shares on Thursday, and 10,000 shares on Friday. The bid depth shows a similar pattern. For all 30 traded stocks, both the bid and ask depth displays a “U-shaped” pattern between Wednesday and Friday, suggesting that market makers are lowering their commitment to sell shares, particularly on Thursday—the day when the insiders were trading. However this pattern does not hold for the no-news sample in Panel (b), where average ask depth increases over these three days. Thus, these univariate results are generally ambiguous as to whether market makers are reacting to informed trading by adjusting ask depth. Additional evidence on the reaction of market makers comes from the results on average quoted and effective spreads. These average quoted spreads show no significant change from one day to the next in both Panels (a) and (b). The average effective spreads show a slight decrease - 13 -
across days. If market makers detected informed trading, then the extant asymmetric information theories of the bid-ask spread would predict that spreads would increase to discriminate against informed traders. These unconditional averages suggest that market makers may not have detected the informed traders, or they may have reacted to them in another manner. Average returns for traded stocks over these 15-minute intervals vary widely across days in Panels (a) and (b). Returns are positive on Wednesday, increase significantly on Thursday, and then are nearly zero on Friday. The Friday results stand out and can be explained by the fact that the information in the IWS column is impounded into the opening price or the first few trades on Fridays. Thus, the intra-day returns show no impact of the IWS column’s release. To measure the degree of buying pressure in the market, we develop a “Buyside” index based on Lee-Ready signed trades. Using the Lee-Ready algorithm, a trade is given the value +1 if it is buyer initiated, –1 if it is seller initiated, and zero if it cannot be signed. These values are summed for all trades in each 15-minute interval to develop the Buyside index value for that interval. As Table 2 shows, buying pressure increased from an average index value of 1.22 on Wednesday to 7.20 on Friday for all traded stocks in Panel (a). The results in Panel (b) show the same pattern of increased demand for these traded stocks. The transaction data results for 44 non-traded stocks are found in Panel (c). Some results are similar to those for traded stocks. Average spreads remain fairly constant across all three days, the volume and number of trades increase, particularly on Friday, and the Buyside index shows increasing buyer interest. Interval volume and Buyside interest show the biggest difference from the earlier results: whereas those two variables increase sharply on Thursday (the insider day) for traded stocks, they actually fall Thursday for the 44 non-traded stocks. In addition, average trade size decreases from 1,998 shares on Wednesday to 1,355 shares on Friday. The relative size of this drop compared to trade size changes in panels (a) and (b) suggests that there was more public interest in these non-traded stocks than in the sample traded by the five stockbrokers. Finally, depth at the ask quotes is increasing and follows a “U-shaped” pattern, which is similar to that found for all 30 traded stocks. - 14 -
C.
Price and Volume Impact Additional information on how the market reacts to stockbroker trading and to the IWS
column is shown in Figures 1 and 2. These figures depict the volume and stock price changes in fifteen-minute intervals, from the open on Wednesday to the close on Friday. They plot, for 21 traded and 44 non-traded stocks with no non-IWS news, the median price and volume changes relative to Wednesday opening prices (Figure 1) and average volumes (Figure 2). Figures 1 and 2
Both Figures 1 and 2 show that trading by the brokers led to increases in volume and price for the affected stocks. Figure 1 shows that, in many of the intervals that follow the onset of insider trading (typically between 1:00 PM and 2:00 PM on Thursdays, indicated by the arrow on the graphs), the median trading volume for those stocks is more than double the average 15minute trading volume on the previous day. In contrast, there is no discernible increase in volume for the stocks mentioned in the IWS column that the brokers did not trade. Consistent with the volume increase, Figure 2 shows a sharp rise in the price of the traded shares, while there is no significant price change for the non-traded stocks. The increase on Thursday occurs mostly after the stockbrokers traded. Thus, supporting the findings of Cornell and Sirri (1992), insiders appear to help start the price discovery process with their trades. The median price increase relative to the average price on Wednesday exceeds 6 percent. Interestingly, the overnight price impact (between Thursday and Friday) of the IWS publication is much stronger for traded stocks (median jump of more than 4 percent) than for non-traded stocks (median jump of just over 2 percent). Figure 2 also shows that, following the large price jump at the open on Friday, there is little price movement during the rest of Friday for traded stocks; in contrast, there is a further 2 percent upward drift for the non-traded stocks.
- 15 -
D.
Returns to Insider Trading Figure 2 suggests that private knowledge of the IWS column may have generated sizable
returns. To investigate this possibility, data are obtained from the Center for Research in Security Prices (CRSP) on the high, low and closing prices of all stocks mentioned in Business Week’s IWS column. For each stock, these data are gathered for a 4-month interval surrounding its mention in the IWS column. The CRSP data are used to estimate a market model and to compute abnormal returns for the stocks, each of the three days surrounding their mention in the IWS column. Those results allow us to determine whether featured stocks experienced abnormal returns, both in the months immediately before, and during the period when, the insiders traded on their advance Business Week information.
IV.
Abnormal Returns After Sant and Zaman’s (1996) results were circulated, the market may have adjusted to
the temporary “bump” in prices generated by the IWS column. In which case, the abnormal returns on Fridays may have diminished or disappeared. This section examines whether abnormal returns existed both before these insiders gained access to the IWS column and during their period of trading.
A.
The Before and During Periods The IWS column first was obtained prior to its public release in June 1995. To examine
abnormal returns before this period, IWS columns are searched for all issues published from November 7, 1994 to May 29, 1995. In those seven months, the IWS column mentioned 107 companies. A total of 26 of these companies are excluded from our analysis: 11 because there is incomplete or missing data in the CRSP files, and 15 because the Dow-Jones News Retrieval service mentioned the company in another news story on Wednesday or Thursday. A sample of 81 companies remains after these exclusions, which is called the “Before” sample. - 16 -
The same procedures are applied to companies mentioned during the period of stockbroker trading. This period covers the IWS columns published between June 5, 1995 and February 5, 1996. There are a total of 116 companies mentioned during this period. A total of 47 companies must be eliminated from the final sample. News articles about the company rules out 38 companies. The remaining 9 companies are excluded because daily CRSP data are incomplete or missing for the estimation period. These eliminations leave a sample of 69 companies, which we refer to as the “During” sample.
B.
An Event Study with Closing Prices Business Week magazine is generally released to newsstands early Friday morning. Some
of the information is available on America Online the night before, but this information is posted after the close of trading in the U.S. Thus, if the information is valuable, its impact on stock prices is expected during trading on Friday. To measure this impact, the event study methodology discussed by Campbell, Lo and MacKinlay (1997) is used for both “Before” and “During” data samples. Specifically, stock returns are computed from the closing price on Thursday (before release of the information) and the closing price on Friday (after the market has had access to the information). These returns are adjusted for market effects by estimating a market model using 90-days of close-to-close returns beginning 10 days before the Wednesday of the announcement week. The Wednesday was chosen as a reference point in order to compute average abnormal returns on Wednesday, Thursday, and Friday. The ten-day gap was used to clearly separate the event being analyzed from the estimates provided by the regression. The market model is estimated using both equal-weighted and value-weighted market indices computed by the CRSP. The results do not change with the choice of index. This procedure was followed for each stock in the sample. Average abnormal returns are computed for Wednesday, Thursday, and Friday of the week that IWS mentioned the company, and two tests are used to determine statistical significance. The J2 test described in Campbell, Lo and MacKinlay (1997) gives better power - 17 -
when the average abnormal return is constant across securities. Given that the potential source of these returns is the same source, this is a reasonable assumption. The second test evaluates the likelihood that more than 50 percent of the abnormal returns in the sample are positive. Table 3 presents the results of these tests. Table 3
Table 3 is divided into two panels. Panel (a) shows the results for the Before sample, which is November 7, 1994 – May 29, 1995. Panel (b) shows the results for the During sample, which includes June 5, 1995 – February 5, 1996. In the Before sample, the average abnormal return is 4.75 percent for Friday (the release day), which is statistically different from zero at the 99-percent level of confidence. There is no evidence of statistically significant abnormal returns for Wednesday or Thursday stock returns. In this sample, 70.4 percent of the abnormal returns are positive, which is also statistically different from 50 percent (the expected level if the column has no effect) at the 99-percent level of confidence. In addition, 75 percent of the “raw” stock returns on Friday are positive for the companies mentioned in the IWS column. In other words, despite the existence of the Sant and Zaman (1996) study, the IWS column still had an influence on the prices of featured stocks. The percent positive abnormal returns are not statistically different from 50 percent for Wednesday or Thursday. The results for the During sample are shown in Panel (b). There is an average abnormal return of 3.87 percent for the public release day, which is statistically significant. In contrast with Panel (a), there are also statistically significant average abnormal returns on Thursday, although at 1.51 percent they are less than one-half the Friday abnormal returns. This result is likely due to leakage of the Business Week information into the market. In this sample, 78.3 percent of the abnormal returns on Friday are positive, which is also statistically significant. In addition, 78 percent of the “raw” stock returns on Friday are positive for companies mentioned in the IWS column. Again, the positive price response on Friday was widespread.
- 18 -
These results show that the stockbrokers could have had a reasonable expectation of profiting from advance access to the IWS column, particularly if their holding period was a single day. As Sant and Zaman (1996) show, the returns from trading on this column are shortlived, so these insiders should be expected to have closed their positions fairly quickly.
C.
The Holding Period The stockbrokers had an incentive to close their positions quickly, rather than risk losing
their gains to market adjustments or mean-reversion after the Business Week information loses its audience. Offsetting this incentive is that rapid turnover may arouse suspicion from exchange authorities or the SEC. Figure 3 shows that these stockbrokers slowly reduced their trading horizon over the eight months that they traded. Figure 3 In the first two months of trading, insiders held the stocks they had purchased for a period of about a week. This period reduces by two days in the next two months, and by the end of the eight-month period to only one and one-half days. Overall, Figure 3 suggests that insiders may have placed less concern on detection after evading discovery for several months and thus began to seek greater profits by shortening their holding period.
V.
Analysis of Stockbroker Trades In this section, we conduct a formal analysis of the impact of the insiders’ trades. The
analysis focuses on stocks traded by the five stockbrokers, with results also provided for a control group of 44 non-traded stocks with no other news on Wednesday or Thursday.
A.
Buying Interest and Interval Returns The abnormal return results show that there is a significant price impact for companies
mentioned in the IWS column. Because the column is publicly available after-hours on - 19 -
Thursday, this impact is generally concentrated at the open on Friday. Even though there is significant buyside interest throughout Friday, the provision of liquidity is high enough that the price impact is negligible after the open. These results are illustrated in Table 4 for all 30 companies traded by the stockbrokers and the 21-company subset that did not have any other news announcements on Wednesday or Thursday. Table 4 For each panel, Table 4 estimates two regression models to explain the buyside index and interval returns using data summarized in 15-minute intervals.16 All regressions are corrected for heteroskedasticity using White’s correction method. The first version (Models 1 and 3 in Panel (a); Models 5 and 7 in Panel (b)) includes dummy variables to capture Thursday and Friday effects relative to Wednesday, which is captured in the constant term. The “Insider Trading Period” dummy variable captures the effects during the intervals in which the stockbrokers were trading. Typically, trading was completed within at most two 15-minute intervals. The “Nasdaq” dummy captures the effect of Nasdaq versus Exchange-listed companies.17 An interaction term is also included to capture the differential effects of insider trading on Nasdaq companies. The second version (Models 2 and 4 in Panel (a); Models 6 and 8 in Panel (b)) omits the “Insider Trading Period” dummy, but includes a dummy variable that captures this period plus the remaining periods in the day. The idea is that this variable captures the effects of other market participants learning of, or reacting to, the informed trading—possibly causing a cascade of buying activity in the market. These other participants may be friends, relatives, customers or associates of the stockbrokers named in the SEC complaint. Also, there may be mimicking or momentum traders noticing the presence of the privately informed traders. Because the “Insider Trading Period” and the “Insider & Remaining Day” variables are highly correlated, they are not 16
Not all 15-minute intervals in a day are included in this sample. An interval is excluded if it contains only zero or one trade. 17 Two companies were listed on the AMEX. These are combined with NYSE companies to form the set of exchange-listed companies.
- 20 -
included in the same regressions. Lastly, an interaction term is included to capture the effects of Nasdaq on the “Insider & Remaining Day” variable. The results from Table 2 (Models 1 and 5) show there is significant buyside interest on both Thursday and Friday relative to Wednesday. Model 1 suggests that buying interest on Friday is over three times the interest on Thursday (6.4 versus 1.9). This diminishes somewhat in Model 5 for the 21 traded stocks without other news (5.4 versus 2.7). Models 2 and 6 show that buyside interest on Thursday is actually confined to the trading intervals following the first insider trade. The earlier part of Thursday shows no significant change relative to Wednesday. Because the “Insider Trading Period” dummy is not significant in these models, one might conjecture that the stockbrokers are successfully disguising their buying interest relative to all order flow in these intervals. However, for the subset of 21 stocks, it appears that the market becomes aware of the higher buying interest after these insiders have completed their trades. These results also show that Nasdaq stocks exhibit significantly higher buying interest than exchange-listed securities. Certainly, there is a preference for Nasdaq stocks by these brokers (55 percent of the companies they traded were Nasdaq-listed). The reason for this may be that the IWS column may have a greater effect on Nasdaq stocks, which are often smaller than exchange listed companies. The regressions for interval returns provide a quantitative summary of the graphical analysis in Figure 2. Figure 2 showed that returns increase toward the end of the day on Thursday and then take a discrete jump between the Thursday close and the open on Friday. After that, Friday returns are volatile, but include a number of periods with negative returns. Models 3, 4, 7, and 8 in Table 4 show that the Thursday dummy variable is not significant. However, the “Insider Period and Remaining Day” dummy variable is significant, which shows that positive returns on Thursday arise after these stockbroker trades. The Friday dummy variable is negative and statistically significant in all of these regressions, which implies that all of the gain from this information is impounded at the open on Friday. The Nasdaq dummy gives consistent results, showing that Nasdaq stocks offer higher returns over these days than - 21 -
exchange-listed stocks. The interaction terms between Nasdaq companies and the time period dummies are not significant. Because the “Insider Trading Period” dummy variable is not significant, these results suggest that it is the mimicking behavior of uninformed traders (or the trades of the brokers’ associates, if they traded after the stockbrokers did) that cause market price impacts. This is a refinement of Meulbroek (1992) and Cornell & Sirri (1992), who show that there are abnormal returns for the day that insiders illegally trade.
B.
Price and Volume Effects To explore further how the brokers’ trades affect the price process, we examine the
number of trades and trade size over these 15-minute intervals. To make results comparable across companies, the data are standardized relative to their respective averages across 15-minute intervals on Wednesday, and then Wednesday is omitted from the analysis. Table 5 presents these regression results using the set of explanatory variables examined in Table 4. Table 5 The relative number of trades is explained in Models 1, 2, 5 and 6. These results show significant increases in trading on Thursday and Friday, with the number of trades on Friday over two and one-half times Wednesday’s trading. The number of trades is higher for Nasdaq versus Exchange-listed, and during the insider trading period. For example, Model 1 indicates that the number of trades is nearly four times greater for Nasdaq stocks during the insider-trading period on Thursday relative to the average on Wednesday. In addition, the number of trades increases relative to Wednesday both during and after insider trading. This is notable because these stockbrokers do not trade a large fraction of the volume on Thursday, which contrasts with Meulbroek’s (1992) findings on trading effects; that is, even a relatively low volume of trading can initiate price effects as shown in Figure 2. The results for relative trade size across traded stocks are explained in Models 3, 4, 7, and 8. Because the regressions for all 30 traded stocks are marginally significant, we will focus this - 22 -
discussion on results for the 21 traded stocks with no news. Models 7 and 8 indicate that relative trade size decreases on Thursday and Friday (coefficient less than one), but increases when only Nasdaq stocks are examined. These results are consistent with the stockbrokers and their followers trading more frequently in smaller sizes on Thursday and the public investors following the same pattern after the news is known on Friday.
C.
Insider Trades and Market Making The key question in this analysis is whether market makers respond to insiders by
changing the bid-ask spreads or the depth offered to the market. Both Cornell and Sirri (1992) and Chakravarty and McConnell (1997) find no significant effect on spreads from actual insider trading. Lee, Mucklow and Ready (1993), Kavajecz (1999) and Koski and Michaely (2000) document that NYSE specialist depth decreases around expected information events, which may signal that informed traders are in the order flow. With a cross-section of companies, we can investigate whether these results generalize beyond NYSE-listed stocks, to specific periods in which traders had valuable information with a limited life that could not be anticipated by the market. Table 6 shows the results of this analysis for quoted spreads, effective spreads, and ask and bid depth using 15-minute interval data. Table 6 As in the analysis of price and volume, the dependent variables in Table 6 are standardized relative to the average value of these variables on Wednesday. Also, the volume of trading (versus Wednesday) is included to control for the effect of volume on spreads and depth. Panel (a) shows the results for all stocks traded by these stockbrokers, and Panel (b) shows the results for the sub-sample of 21 companies without any other news. The spread results are shown in Models 1 to 4 and Models 9 to 12. For Panel (a), the coefficients on the Thursday and Friday dummy variables show that both quoted and effective spreads are slightly lower on these days than their average value on Wednesday. This result, however, only holds for quoted spreads in - 23 -
Panel (b), as effective spreads are nearly equal to the averages reported on Wednesday. These results also depend on whether the company is Nasdaq- or exchange-listed. When the Thursday and Friday dummies are combined with the Nasdaq dummy, the spreads are not statistically different than on Wednesday in both panels. Importantly, quoted spreads show no response to trading by these informed stockbrokers in both panels; only effective spreads in Panel (a) show a slight significant negative response. The latter, of course, is not the anticipated response of market makers who detect informed traders; that is, they would increase not decrease spreads. These results generalize the findings of Cornell and Sirri (1992) and Chakravarty and McConnell (1997) to a larger number of Nasdaq- and Exchange-listed stocks. The only evidence that insiders may be affecting spreads comes from the effective spread regressions for all 30 stocks. Similar to the startling results in Cornell and Sirri, this evidence suggests that, if anything, informed trades possibly face slightly lower execution costs. Table 6 also shows how insider trading affects market depth. Models 5, 6, 13 and 14 reveal that insider trades significantly affect depth on the ask side. Because these insiders were buying stock, ask depth is the side most likely affected. Models 7, 8, 15 and 16 confirm that these trades do not affect depth on the bid side. The results on the ask side offer several insights into how market makers adjust to insiders. Models 5 and 13 show that ask depth is lower during the insider period. Hence, even though the bid-ask spread does not generally change, the market maker reduces its risk by offering a smaller quantity at the posted price. This extends to episodes in which insider trading is unexpected, using data in which informed trading is exactly identified and using both Nasdaq and NYSE stocks, the results of Lee, Mucklow and Ready (1993), Kavajecz (1999) and Koski and Michaely (2000) that market makers use quoted depth to manage asymmetric information risk. Note that the depth decrease on the ask side is not dependent on whether it is a Nasdaq or Exchange-listed stock; depth decreases for both trading venues. The results in Models 5 and 13 show that the Nasdaq/Insider Period interaction term is positive and that its size is larger than the Nasdaq dummy variable effect. As such, Nasdaq depth does not decrease as much as NYSE - 24 -
depth during the insider-trading period measured relative to average depth on Wednesday. This result is understandable in a dealer versus specialist market because the diffuse nature of a dealer market makes it more difficult for a given dealer to determine the information content of the order flow. Thus, it is more difficult for a given dealer to avoid being “picked off” by informed traders. The results for ask depth with the intervals during and after insider trading combined are mixed in Models 6 and 14. Model 6 (all 30 traded stocks) shows that depth increases following the insider-trading period, but this dummy variable is not significant in Model 14 (21 traded stocks with no news). Because the fit is significantly better in Model 14, this regression is likely more conclusive. As such, we may conclude that ask depth decreases during the insider-trading period, but rebounds later in the day, possibly because the price moves have attracted other liquidity providers to the market or because the dealers/specialists have returned after the insiders have departed.
D.
A Comparison with Non-traded Stocks These stockbrokers chose to trade only selected stocks mentioned in the Business Week
column. We know that these stocks are different from the stocks not traded in size and listing venue, but the question remains as to whether they are different in market making characteristics. To explore this question, regressions are estimated for 44 non-traded companies that had no other news announcement on Wednesday or Thursday. Two “hypothetical” dummy variables are constructed to capture the time that the insiders would most likely have traded these stocks. This variable is one during the period 1:00 to 1:30 PM in the first set of regressions and zero otherwise. In the second set of regressions, the time period slides one-half hour to 1:30 to 2:00 PM. Table 7 presents the results of this analysis. Table 7
- 25 -
Table 7 reports regressions for the number of trades, trade size, quoted and effective spreads, and bid and ask depth. These variables are all measured relative to their average values on Wednesday. The results for the Thursday and Friday dummy variables in the number of trades regressions are similar to those in Table 5; that is, there are relatively more trades on these days. Interestingly, there are significantly fewer trades during the hypothetical insider trading periods. In addition, the Nasdaq dummy variable is not significant. The results for trade size contrast with Table 5 in that trade sizes for these stocks tend to be relatively higher on Thursday and Friday. None of the other variables are significant in the trade size regressions. The quoted and effective spread results are similar to those reported in Table 6, except that quoted spreads are significantly lower for these Nasdaq stocks versus the exchange-listed stocks. Importantly, the hypothetical insider trading period dummy variables are not significant, so that there is not anything unusual about spreads on these non-traded stocks during the time that the stockbrokers were in the market. Finally, the depth regressions confirm that depth is not changing for these stocks during the hypothetical insider periods. In general, the results for the non-traded stocks are a good control group for the effects that may influence market making in the stocks mentioned by Business Week. That there is no indication that ask depth is changing for these non-traded stocks further increases confidence that the relative decrease in ask depth for the trading stocks is due to these stockbroker trades.
E.
Unbundling Liquidity Providers The data available for this study do not show who is trading with whom. As such, it is not
possible to separate liquidity providers into market makers and limit order traders to determine which group is adjusting most to these informed trades. It is possible that the informed trades exhaust the inside limit orders and that market makers are left quoting their own commitment, which may be unchanged. In this event, spreads may not change if market makers (particularly specialists) have not detected the informed trading, but depth will decrease. Although we cannot
- 26 -
completely rule out this possibility, the evidence between dealer and specialist markets makes it less likely. For these stockbrokers, the average trade size is 5,320 shares in Nasdaq stocks and 4,683 shares in exchange-listed stocks. The average depth at the ask is 2,809 shares on Nasdaq and 16,395 shares on the exchanges. With depth for exchange-listed stocks about 5.8 times the depth of Nasdaq stocks and the stockbrokers’ trade size is smaller on the exchanges, we would expect a greater reaction in depth on the Nasdaq if the stockbrokers’ trades were only exhausting inside limit orders on the book. However, Table 6 shows that depth on the exchanges reacts more than depth on the Nasdaq. Thus, it appears that specialists on the exchanges are playing an active role in managing quoted depth during these insider periods.18 This view is also consistent with Kavajecz (1999), who finds that depth provided by both specialists and limit orders decreases prior to earnings announcements for NYSE stocks.
VI.
Conclusions Using a unique episode of repeated insider trading by outsiders across a group of stocks,
we show that liquidity providers do not adjust spreads during periods of genuinely informed trading. Instead, they adjust offered depth to reduce the risk of transacting with informed traders. This result holds regardless of the type of market where companies traded, although the magnitude of the depth decrease is less for Nasdaq-listed stocks compared to exchange-listed companies. In addition, we show that, during and immediately following periods when insiders are buying shares, trades are much more numerous—yet smaller in size—than at other times. The results show that trades during those periods are overwhelmingly buyer-initiated. In contrast to earlier findings, however, insiders’ trades do not account for the major fraction of the trading volume increase. Our evidence suggests that the volume increases may reflect an increase in noise trading by “falsely informed” or mimicking traders. 18
The median results provide the same conclusions. The median trade size by stockbrokers is 4,000 shares in Nasdaq stocks and 3,500 shares in exchange-listed stocks. The median depth is 2,732 on Nasdaq and 10,915 on the exchanges.
- 27 -
References Barber, Brad M. and Douglas Loeffler, 1993, “The ‘Dartboard’ Column: Second-Hand Information and Price Pressure,” Journal of Financial and Quantitative Analysis, 28 (2), pp. 273-84. Barclay, Michael J., 1997, “Bid-Ask Spreads and the Avoidance of Odd-Eighth Quotes on Nasdaq: An Examination of Exchange Listings,” Journal of Financial Economics, 45 (1), pp.35-60 Barclay, Michael J., William G. Christie, Jeffrey H. Harris, E. Kandel and Paul H. Schultz, 1999, “Effects of Market Reform on the Trading Costs and Depths of Nasdaq Stocks,” Journal of Finance, 54 (1), pp.1-34. Beneish, Messod D., 1991, “Stock Prices and the Dissemination of Analysts' Recommendations,” Journal of Business, 64 (3), pp. 393-416. Bessembinder, Hendrik, 1997, “The Degree of Price Resolution and Equity Trading Costs,” Journal of Financial Economics, 45 (1), pp. 9-34. Bessembinder, Hendrik, 1999, “Trade Execution Costs on Nasdaq and the NYSE: A Post-reform Comparison,” Journal of Financial and Quantitative Analysis, 34 (3), pp.387-407. Bessembinder, Hendrik and Herbert M. Kaufman, 1997, “A Cross-Exchange Comparison of Execution Costs and Information Flow for NYSE-Listed Stocks,” Journal of Financial Economics, 46 (3), pp. 293-319. Bessembinder, Hendrik and Herbert M. Kaufman, 1997, “A Comparison of Trade Execution Costs for NYSE and Nasdaq-Listed Stocks,” Journal of Financial and Quantitative Analysis, 32 (3), pp. 287310. Chakravarty, Sugato and John J. McConnell, 1997, “An Analysis of Prices, Bid/Ask Spreads, and Bid and Ask Depths Surrounding Ivan Boesky's Illegal Trading in Carnation's Stock,” Financial Management, 26 (2), pp. 18-34. Chakravarty, Sugato and John J. McConnell, 1999, “Does Insider Trading Really Move Stock Prices?” Journal of Financial and Quantitative Analysis, 34 (2), pp. 191-209. Christie, William G., Shane Corwin, and Jeffrey H. Harris, 2002, “Nasdaq Trading Halts: The Impact of Market Mechanisms on Price, Trading Activity and Execution Costs,” forthcoming in the Journal of Finance. Chung, Kee H., Bonnie F. Van Ness and Robert .A. Van Ness, 2001, “Can the Treatment of Limit Orders Reconcile the Differences in Trading Costs between NYSE and Nasdaq Issues?” Journal of Financial and Quantitative Analysis, 36 (2), pp. 267-86. Clyde, Paul, Paul Schultz and Mir Zaman, 1997, “Trading Costs and Exchange Delisting: The Case of Firms That Voluntarily Move from the American Stock Exchange to the Nasdaq,” Journal of Finance, 52 (5), pp. 2103-12. Cornell, Bradford and Erik R. Sirri, 1992, “The Reaction of Investors and Stock Prices to Insider Trading,” Journal of Finance, 47 (3), pp. 1031-59.
- 28 -
Corwin, Shane A. and Marc L. Lipson, 2000, “Order Flow and Liquidity around NYSE Trading Halts,” Journal of Finance, 55 (4), pp. 1771-1801. Dupont, Dominique, 2000, “Market making, Prices, and Quantity Limits,” Review of Financial Studies, 13 (4), pp. 1129-51. Easley, David and Maureen O’Hara, 1992, “Time and the Process of Security Price Adjustment,” Journal of Finance, 57 (2), pp. 577-605. Foster, F. Douglas and S. Vishwanathan, 1995, “Trading Costs for Target Firms Around Takeovers.” In Hirschey, Mark & M. Wayne Marr (Eds.), Advances in Financial Economics, JAI Press, Greenwich, Conn., pp.37-57. Garfinkel, Jon A. and M. Nimalendran, 2001, “Market Structure and Trader Anonymity: An Analysis of Insider Trading,” Working Paper, University of Iowa and University of Florida, November. Glosten, Lawrence R., 1989, “Insider Trading, Liquidity and the Role of the Monopolist Specialist,” Journal of Business, 62 (2), pp. 211-35. Glosten, Lawrence R. and Paul R. Milgrom, 1985, “Bid, Ask and Transaction Prices in a Specialist Market with Heterogeneously Informed Traders,” Journal of Financial Economics, 14, pp. 71-100. Greene, Jason and Scott Smart, 1999, “Liquidity Provision and Noise Trading: Evidence from the ‘Investment Dartboard’ Column,” Journal of Finance, 54 (5), pp. 1885-99. Heidle, Hans G. and Roger D. Huang, 2002, “Information-Based Trading in Dealer and Auction Markets: An Analysis of Exchange Listings” forthcoming in the Journal of Financial and Quantitative Analysis. Huang, Roger D. and Hans R. Stoll, 1996, “Dealer versus Auction Markets: A Paired Comparison of Execution Costs on Nasdaq and the NYSE,” Journal of Financial Economics, 41 (3), pp. 313-57. Jennings, Robert, 1994, “Intraday Changes in Target Firms’ Share Price and Bid-Ask Quotes around Takeover Announcements,” Journal of Financial Research, 17 (2), pp. 255-70. Kavajecz, Kenneth, 1998, “A Specialist’s Quoted Depth as a Strategic Choice Variable,” Working Paper, The Wharton School of the University of Pennsylvania, May. Kavajecz, Kenneth, 1999, “A specialist’s Quoted Depth and the Limit Order Book,” Journal of Finance, 54 (2), pp. 747-71. Kim, Oliver and Robert E. Verrecchia, 1994, “Market Liquidity and Volume around Earnings Announcements,” Journal of Accounting and Economics, 17, (1-2), pp. 41-67. Koski, Jennifer Lynch and Roni Michaely, “Prices, Liquidity, and the Information Content of Trades,” Review of Financial Studies, 13 (3), pp. 659-96. Krinsky, Itzhak and Jason Lee, 1996 “Earnings Announcements and the Components of the Bid-Ask Spread,” Journal of Finance, 51 (4), pp. 1523-35. Kyle, Albert S., 1985, “Continuous Auctions and Insider Trading,” Econometrica, 53 (6), pp. 1315-35.
- 29 -
LaPlante, Michele and Chris J. Muscarella, 1997, “Do Institutions Receive Comparable Execution in the NYSE and Nasdaq Markets? A Transaction Study of Block Trades,” Journal of Financial Economics, 45 (1), pp. 97-134. Lee, Charles M. C. and Mark Ready, 1991, “Inferring Trade Direction from Intraday data,” Journal of Finance, 45, pp. 733-46. Lee, Charles M. C., Belinda Mucklow and Mark J. Ready, 1993, “Spreads, Depths, and the Impact of Earnings Information: An Intraday Analysis,” Review of Financial Studies, 6 (2), pp. 345-74. Liang, Bing, 1999, “Price Pressure: Evidence from the ‘Dartboard’ Column,” Journal of Business, 72 (1), pp. 119-34. Liu, Pu, Stanley D. Smith and Azmat A. Syed, 1990, “Stock Price Reactions to the Wall Street Journal’s securities recommendations, ” Journal of Financial and Quantitative Analysis, 25 (3), pp. 399-410. Lloyd-Davies, Peter and Michael Canes, 1979, “Stock Prices and the Publication of Second-Hand Information,” Journal of Business, 51 (1), pp. 43-56. Madhavan, Ananth, 2000, “Market Microstructure: A Survey,” Journal of Financial Markets, 3, pp. 20558. Meulbroek, Lisa K., 1992, “An Empirical Analysis of Illegal Insider Trading,” Journal of Finance, 47 (5), pp. 1661-99. Odean, Terrance, 1998, “Volume, volatility, price, and profit when all traders are above average,” Journal of Finance, 53 (6), pp. 1887-1934. Pound, John A. and Richard J. Zechauser, 1990, “Clearly Heard on the Street: The Effect of Takeover Rumors in Stock Prices,” Journal of Business, 63 (3), pp. 291-308. Sant, Rajiv and Mir A. Zaman, 1996, “Market Reaction to Business Week ‘Inside Wall Street’ Column: A Self-Fulfilling Prophecy,” Journal of Banking and Finance, 20 (4), pp. 617-43. Stoll, Hans R., 2000, “Friction,” Journal of Finance, 55 (4), pp. 1479-1514. Weston, James P., 2000, “Competition on the Nasdaq and the Impact of Recent Market Reforms,” Journal of Finance, 55 (6), pp. 2565-98. Wijmenga, Roel T., 1990, “The Performance of Published Dutch Stock Recommendations,” Journal of Banking and Finance, 14, pp. 559-81.
- 30 -
Table 1 Characteristics of Traded and Non-traded Companies Mentioned by Business Week Summary data are presented for 30 companies traded by stockbrokers and 76 not-traded companies mentioned in Business Week's "Inside Wall Street" column. The Dow Jones News Retrieval Service was searched for mention of these stocks in another news source. The financial data are from Compustat. Compustat does not cover all of the firms in our sample. Specifically, the financial characteristics are for 20 traded firms and 60 non-traded firms. Generally, the omitted firms are smaller, which gives an upward bias to sales and total assets information. Traded
Non-traded
Business Week Sentiment Count Percent
Buy 30 100.0%
Sell 0 0.0%
Buy 72 94.7%
Sell 4 5.3%
Mentioned in Another News Source? Count Percent
Yes 9 30.0%
No 21 70.0%
Yes 22 28.9%
No 54 71.1%
Exchange Listed Count Percent
NYSE / AMEX 12 / 2 45.2%
Nasdaq 16 54.8%
NYSE / AMEX 38 / 6 57.9%
Nasdaq 32 42.1%
Rate of Return on Assets Median (Average) Standard Deviation
1994 1995 1.6% (-3.5%) 1.8% (-0.8%) 18.4% 10.1%
1994 3.6% (3.1%) 12.5%
1995 3.8% (-1.2%) 44.0%
Rate of Return on Equity Median (Average) Standard Deviation
4.7% (-1.1%) 24.8%
Sales (millions of dollars) Median (Average) Standard Deviation
218.3 (1529.9) 364.3 (1713.4) 3604.1 4066.6
Total Assets (millions of dollars) Median (Average) Standard Deviation
374.2 (2294.6) 357.7 (2401.8) 6055.6 6220.0
413.9 (8825.8) 406.5 (8978.4) 29279.4 32960.4
Growth Rate of Sales (1994-1995) Median (Average) Standard Deviation
17.5% (33.2%) 58.9%
21.3% (32.7%) 49.1%
Growth Rate of Net Income (1994-95) Median (Average) Standard Deviation
-58.3% (280.8%) 1377.2%
16.6% (-26.6%) 416.9%
3.6% (4.6%) 17.6%
11.2% (13.8%) 11.5% (-22.4%) 55.2% 221.2% 261 (3022.2) 8925.1
297.3 (3383.2) 10184.0
Table 2 Descriptive Statistics of the Transaction Data This table presents the average and median values for variables and samples examined in this study. These data are computed for all trading days and for Wednesday, Thursday, and Friday separately. The transaction data are summarized into 15-minute intervals with the average shown computed across these intervals. Panel (a) summarizes information for all 30 stocks traded by the five brokers with access to advance copies of BusinessWeek magazine. Panel (b) is for a sub-set of 21 traded stocks that had no other news announced on Wednesday or Thursday. Panel (c) shows information for 44 non-traded companies mentioned in BusinessWeek that did not have any other news on Wednesday or Thursday. Missing transaction data for an additional four non-traded, no-news companies ruled out their inclusion in Panel (c).
All Days Variable
Mean
Wednesday
Median
Mean
Median
Thursday Mean
Median
Friday Mean
Median
Panel (a): 30 Stocks Traded by Insiders Stock Price Quoted Spread Effective Spread Bid Depth (100s) Ask Depth (100s) Trade Count Trade Size Interval Volume Interval Returns (%) Buyside Index
19.75 0.25 0.13 65 89 12.6 1664 20147 0.1021 3.80
14.50 0.20 0.10 35 38 7.0 1589 9500 0.0233 2.25
Stock Price Quoted Spread Effective Spread Bid Depth (100s) Ask Depth (100s) Trade Count Trade Size Interval Volume Interval Returns (%) Buyside Index
17.93 0.28 0.15 46 60 11.6 1712 19979 0.1469 3.90
13.13 0.22 0.11 27 32 6.0 1618 9696 0.0496 2.75
19.26 0.25 0.15 66 86 8.3 1735 14153 0.0831 1.22
13.42 0.19 0.11 34 41 5.0 1591 7200 0.0000 1.11
19.55 0.25 0.13 63 80 10.5 1704 18345 0.2139 3.09
14.00 0.19 0.10 40 35 7.0 1679 9000 0.0998 2.54
20.62 0.26 0.12 66 100 17.1 1550 25616 0.0063 7.20
15.22 0.21 0.09 34 39 9.0 1519 11854 -0.0312 4.15
18.57 0.28 0.13 54 76 15.1 1615 25610 -0.0132 6.50
13.35 0.23 0.10 29 36 8.0 1577 11904 -0.0377 4.61
23.08 0.27 0.13 71 81 23.2 1355 35490 0.0727 8.07
17.28 0.21 0.09 10 15 11.0 1364 12200 0.0676 3.10
Panel (b): 21 Traded Stocks Without Any Other News 17.39 0.28 0.16 42 51 6.7 1751 11005 0.1044 1.45
12.62 0.23 0.13 27 31 4.0 1393 6000 0.0000 1.13
17.83 0.27 0.15 40 54 10.8 1771 19445 0.3494 3.74
13.13 0.21 0.11 26 28 6.0 1728 9900 0.2598 3.33
Panel (c): 44 Non-Traded Stocks Without Any Other News Stock Price Quoted Spread Effective Spread Bid Depth (100s) Ask Depth (100s) Trade Count Trade Size Interval Volume Interval Returns (%) Buyside Index
22.63 0.26 0.14 68 73 17.1 1584 29150 0.0576 4.14
16.93 0.20 0.09 10 14 8.0 1305 9605 0.0020 1.60
22.36 0.26 0.14 60 74 13.2 1998 27529 0.0328 2.30
16.65 0.19 0.10 11 15 7.0 1437 8602 0.0000 1.15
22.45 0.25 0.15 71 64 12.9 1399 22126 0.0673 2.06
16.71 0.20 0.10 10 13 7.0 1165 8498 -0.0043 1.15
Table 3 Average Abnormal Returns on Wednesday, Thursday, and Friday (Release Day) for Companies Mentioned in Business Week's "Inside Wall Street" Column Average abnormal returns are computed using a market model for stocks mentioned in Business Week between November 7, 1994 to May 29, 1995, which is before the Inside Wall Street column became available ahead of publication, and for the period June 5, 1995 to January 29, 1996, which is during the period of insider trading. Stock with other news announcements on Wednesday or Thursday are excluded from these samples. The results shown use the equal-weighted CRSP index to measure overall market returns. Results that are statistically significant at the 99-percent level of confidence are shown with an "**". Wednesday
Thursday
Friday
Panel (a): "Before" Period Sample of 81 Companies Average Abnormal Return Asymptotic Normal J2 Test Percent Positive Abnormal Returns Asymptotic Normal Z-test
0.64%
0.27%
4.75%
1.62
0.68
12.05**
53.09% 0.56
56.79% 1.22
70.31% 3.67**
Panel (b): "During" Period Sample of 69 Companies Average Abnormal Return Asymptotic Normal J2 Test Percent Positive Abnormal Returns Asymptotic Normal Z-test
0.77%
1.51%
3.87%
1.99
3.90**
10.01**
46.38% -0.61
47.83% -0.36
78.26% 4.69**
Table 4 Buying Sentiment and Interval Returns on Wednesday, Thursday and Friday Regressions are run for 30 stocks traded by brokers for all companies combined (Panel a) and excluding 9 companies that had other news announcements on Wednesday or Thursday (Panel b). Transactions data are analyzed in 15 minute intervals on Wednesday, Thursday and Friday. Trades are signed with the Lee-Ready algorithm (+1 for buyer initiated and -1 for seller initiated). The "Buyside Index" measures buying sentiment as the sum of the signed trades in the interval. The larger the sum, the more buyer-initiated trades. "Interval Returns" are computed from the last trade in the previous interval to the last trade in the present interval. The independent variables are as follows. "Thursday" and "Friday" are dummy variables for these trading days. "Insider Trading Period" is a dummy variable for intervals of insider trading. "Insider Period and Remaining Day" equals one on Thursday for all intervals after the first insider trade. The "Nasdaq" dummy equals one for Nasdaq stocks; zero for exchange-listed stocks. The two interaction terms measure the effect of insider trading on Nasdaq stocks. Regressions are corrected for heteroscedasticity using White's method. p-values are shown in parentheses below the coefficients. Panel (a): 30 Stocks Traded by Insiders
Panel (b): 21 Traded Stocks Without Any Other News
Buyside Index Model 1 Model 2
Interval Returns Model 3 Model 4
Buyside Index Model 5 Model 6
Interval Returns Model 7 Model 8
Constant
-0.077 (0.851)
0.012 (0.976)
0.001 (0.115)
0.001 (0.052)
0.655 (0.156)
0.997 (0.039)
0.001 (0.269)
0.001 (0.159)
Thursday
1.936 (0.000)
0.204 (0.667)
0.000 (0.846)
-0.001 (0.160)
2.747 (0.000)
0.519 (0.356)
0.001 (0.538)
-0.001 (0.402)
Friday
6.423 (0.000)
6.443 (0.000)
-0.002 (0.008)
-0.002 (0.009)
5.389 (0.000)
5.408 (0.000)
-0.003 (0.011)
-0.003 (0.011)
Insider Trading Period
2.608 (0.300)
Variables
Insider Period and Remaining Day
3.299 (0.149)
Nasdaq Dummy
4.752 (0.000)
Nasdaq*Insider Period
-0.093 (0.761)
Nasdaq*Insider & Remaining Day
Adjusted R-Squared F-test of Regression Observations
0.005 (0.175)
4.553 (0.000)
0.003 (0.013) 0.002 (0.002)
0.002 (0.013)
0.001 (0.898) 1.801 (0.229)
0.086 29.34 1515
4.596 (0.399)
0.094 32.49 1515
2.158 (0.023) 2.803 (0.000)
2.266 (0.000)
-2.136 (0.710) 0.003 (0.145)
0.013 4.94 1515
0.012 (0.145)
0.019 6.71 1515
0.003 (0.080) 0.003 (0.003) -0.006 (0.525)
4.465 (0.006) 0.056 12.86 996
0.002 (0.018)
0.076 17.42 996
0.003 (0.207) 0.018 4.68 996
0.021 5.34 996
Table 5 Effects of Insider Trades on Number of Trades and Trade Size Regressions are computed for 30 companies traded by insiders using transactions data analyzed in 15 minute intervals on Wednesday, Thursday (the insider trading day) and Friday. Panel (a) shows the results with all 30 stocks and Panel (b) shows results excluding 9 stocks that had other news announcements on either Wednesday or Thursday. All dependent variables are relative to the average over 15-minute intervals on Wednesday. "Number of Trades" is the total number of transactions during the interval and "Trade Size" is the average volume of shares traded during the interval. These dependent variables are measured relative to their average on Wednesday, the day before any inside information was obtained. The independent variables are defined as in Table 2. All regressions are corrected for heteroscedasticity using White's correction method. p-values are shown in parentheses below each estimated coefficient. Panel (a): 30 Stocks Traded by Insiders
Panel (b): 21 Traded Stocks Without Any Other News
Number of Trades Model 1 Model 2
Trade Size Model 3 Model 4
Number of Trades Model 5 Model 6
Trade Size Model 7 Model 8
Thursday
1.068 (0.000)
0.668 (0.000)
0.960 (0.000)
0.906 (0.000)
1.603 (0.000)
1.155 (0.000)
0.867 (0.000)
0.819 (0.000)
Friday
2.670 (0.000)
2.673 (0.000)
0.894 (0.000)
0.920 (0.000)
2.683 (0.000)
2.790 (0.000)
0.781 (0.000)
0.816 (0.000)
Insider Trading Period
1.747 (0.067)
Variables
Insider Period and Remaining Day
1.028 (0.000)
Nasdaq Dummy
1.188 (0.000)
Nasdaq*Insider Period
-0.129 (0.932)
Nasdaq*Insider & Remaining Day
Adjusted R-Squared F-test of Regression Observations
0.437 (0.093)
1.183 (0.000)
0.106 (0.294) 0.119 (0.038)
0.071 (0.255)
-0.315 (0.298) 0.204 (0.674)
0.070 21.61 1101
3.754 (0.048)
0.074 23.05 1101
0.771 (0.012) 0.371 (0.061)
0.213 (0.357)
-1.666 (0.470) 0.226 (0.119)
0.005 2.31 1101
0.080 (0.757)
0.010 3.68 1101
0.019 (0.852) 0.314 (0.000) -0.068 (0.827)
1.065 (0.048) 0.044 9.45 739
0.263 (0.000)
0.051 10.92 739
0.252 (0.086) 0.025 5.64 739
0.032 7.03 739
Table 6 Effects of Insider Trades on Market Makers Spreads and Depth Regressions are computed for 30 stocks traded by insiders using transactions data analyzed in 15 minute intervals on Wednesday, Thursday and Friday. Panel (a) shows the results with all 30 stocks while Panel (b) shows results excluding 9 stocks that had other news announcements on either Wednesday or Thursday. "Quoted Spread" is the average bid-ask spread during the interval, "Effective Spread" equals two times the absolute value of the difference between price and the midpoint of the bid-ask spread, averaged over the interval. "Depth" is reported at the best bid and offer averaged over the interval. For Nasdaq stocks, depth is aggregated across all market makers quoting at the best bid or ask. All dependent variables are measured relative to their average over 15-minute intervals on Wednesday, the day before the inside information was obtained. The independent variables are all defined in Table 2, except "Relative Volume", which equals the volume in the trading interval relative to the average volume across all 15 minute intervals on Wednesday. White's method is used to correct for heteroscedasticity. p-values are shown in parentheses below each estimated coefficient.
Panel (a): 30 Stocks Traded by Insiders Quoted Spread Model 1 Model 2
Effective Spread Model 3 Model 4
Ask Depth Model 5 Model 6
Bid Depth Model 7 Model 8
Thursday
0.893 (0.000)
0.913 (0.000)
0.972 (0.000)
0.976 (0.000)
1.259 (0.000)
1.096 (0.000)
1.208 (0.000)
1.127 (0.000)
Friday
0.890 (0.000)
0.885 (0.000)
0.995 (0.000)
0.987 (0.000)
1.732 (0.000)
1.724 (0.000)
1.213 (0.000)
1.211 (0.000)
Insider Trading Period
-0.027 (0.586)
Variables
Insider Period and Remaining Day
-0.061 (0.034) -0.041 (0.291)
Nasdaq Dummy
0.082 (0.001)
Nasdaq*Insider Period
-0.156 (0.176)
Nasdaq*Insider & Remaining Day
0.088 (0.001)
-0.634 (0.000) -0.002 (0.942)
0.031 (0.052)
0.044 (0.011)
-0.015 (0.767) -0.067 (0.239)
-0.084 (0.678) 0.326 (0.074)
-0.437 (0.000)
-0.402 (0.004)
0.785 (0.003) -0.072 (0.045)
0.181 (0.248) -0.380 (0.000)
-0.368 (0.000)
0.115 (0.688) -0.022 (0.917)
-0.019 (0.910)
Relative Volume
0.048 0.008
0.051 0.006
0.053 0.004
0.054 0.003
0.099 0.247
0.087 0.310
0.169 0.001
0.164 0.001
Adjusted R-Squared F-test of Regression Observations
0.020 5.42 1101
0.021 5.75 1101
0.039 9.89 1101
0.042 10.64 1101
0.026 6.95 1101
0.026 6.95 1101
0.051 12.86 1101
0.054 13.63 1101
Panel (b): 21 Traded Stocks Without Any Other News Quoted Spread Model 9 Model 10 Thursday
Effective Spread Model 11 Model 12
Ask Depth Model 13 Model 14
Bid Depth Model 15 Model 16
0.849 0.000
0.833 0.000
0.990 0.000
1.015 0.000
1.737 0.000
1.760 0.000
1.324 0.000
1.181 0.000
Friday
0.921 (0.000)
0.907 (0.000)
1.032 (0.000)
1.029 (0.000)
2.132 (0.000)
2.190 (0.000)
1.355 (0.000)
1.338 (0.000)
Insider Trading Period
0.035 (0.678)
Insider Period and Remaining Day
0.047 (0.394)
Nasdaq Dummy
0.149 (0.000)
Nasdaq*Insider Period
-0.189 (0.184)
Nasdaq*Insider & Remaining Day
Relative Volume
Adjusted R-Squared F-test of Regression Observations
-0.040 (0.325)
0.168 (0.000)
-1.327 (0.000) -0.040 (0.329)
0.245 (0.214)
0.026 (0.265)
-0.206 (0.733) -0.105 (0.114)
-0.366 (0.190) -0.214 (0.520)
-1.188 (0.000)
-1.264 (0.000)
1.591 (0.000) -0.036 (0.351)
0.230 (0.310) -0.564 (0.000)
-0.531 (0.000)
0.474 (0.185) 0.477 (0.176)
0.011 (0.964)
0.021 (0.354)
0.023 (0.321)
0.048 (0.007)
0.050 (0.005)
0.272 (0.046)
0.263 (0.053)
0.208 (0.006)
0.202 (0.007)
0.031 5.69 739
0.031 5.71 739
0.029 5.33 739
0.033 6.09 739
0.074 12.76 739
0.073 12.56 739
0.070 12.03 739
0.074 12.71 739
Table 7 Comparison of "Insider Effects" for Non-Traded Stocks Regressions are computed for 44 companies, mentioned in Business Week issues from June 15 1995 to February 6 1996 but not traded by insiders. These companies did not have any other news announcement on either Wednesday or Thursday, the day the inside information was obtained. Transactions data are analyzed in 15 minute intervals on Thursday and Friday, with the dependent variables relatives to the average on Wednesday. The "Hypothetical Insider Period" dummy is set equal to unity in two periods, 1:00 to 1:30 PM and 1:30 to 2:00 PM, which correspond to the period that insiders traded other stocks mentioned by Business Week . The results are shown for each period separately. Dependent and independent variables are as defined in previous tables. White's method is used to correct for heteroscedasticity. p-values are shown in parentheses below each estimated coefficient. All Dependent Variables are Relative to the Average Over 15 Minute Intervals on Wednesday Variables
Number of Trades 1:00-1:30 1:30-2:00
Trade Size 1:00-1:30 1:30-2:00
Quoted Spread 1:00-1:30 1:30-2:00
Effective Spread 1:00-1:30 1:30-2:00
Ask Depth 1:00-1:30 1:30-2:00
Bid Depth 1:00-1:30 1:30-2:00
Thursday
1.100 (0.000)
1.101 (0.000)
1.125 (0.000)
1.124 (0.000)
0.963 (0.000)
0.963 (0.000)
0.946 (0.000)
0.945 (0.000)
1.054 (0.000)
1.023 (0.000)
0.926 (0.000)
0.909 (0.000)
Friday
2.309 (0.000)
2.312 (0.000)
1.078 (0.000)
1.069 (0.000)
0.998 (0.000)
0.999 (0.000)
0.962 (0.000)
0.962 (0.000)
1.059 (0.000)
1.049 (0.000)
0.934 (0.000)
0.930 (0.000)
Hypothetical Insider Period
-0.373 (0.002)
-0.307 (0.005)
-0.275 (0.274)
-0.036 (0.897)
-0.361 (0.594)
-0.045 (0.672)
0.049 (0.274)
0.035 (0.663)
0.214 (0.493)
0.669 (0.183)
0.327 (0.233)
0.472 (0.295)
Nasdaq Dummy
0.030 (0.762)
0.023 (0.816)
-0.134 (0.062)
-0.116 (0.111)
-0.049 (0.015)
-0.052 (0.009)
0.036 (0.003)
0.035 (0.004)
-0.179 (0.000)
-0.157 (0.001)
-0.241 (0.000)
-0.233 (0.000)
Nasdaq*Insider Period
0.214 (0.369)
0.292 (0.371)
0.487 (0.169)
-0.232 (0.432)
0.022 (0.851)
0.100 (0.540)
-0.061 (0.466)
0.008 (0.933)
-0.070 (0.847)
-0.591 (0.240)
-0.327 (0.350)
-0.410 (0.369)
0.043 (0.004)
0.043 (0.004)
0.045 (0.000)
0.045 (0.000)
0.158 (0.001)
0.158 (0.001)
0.435 (0.010)
0.435 (0.010)
0.028 9.63 1510
0.028 9.73 1510
0.071 24.1 1510
0.071 24.16 1510
0.054 18.38 1510
0.061 20.73 1510
0.196 74.72 1510
0.198 75.36 1510
Relative Volume
Adjusted R-Squared F-test of Regression Observations
0.093 39.78 1510
0.093 39.74 1505
0.000 1.06 1510
0.000 0.91 1510
Figure 1 Three-Day Median Volume Changes This figure shows the median percentage trading volume from the open on the Wednesday preceding the release of the relevant IWS column until the close on the Friday when the magazine is publicly released. These data are measured in 15-minute intervals relative to the average volume on Wednesday. This plot is drawn for the 21 stocks that at least one of the brokers traded (“traded”) and 44 that no insider traded according to the SEC complaint (“not traded”), using transactions data summarized in 15minute intervals. Only stocks not mentioned in another news source on the insider trading day (Th) or the day before (W) are included. The two vertical lines represent the end of the first (W) and second (Th) trading days. The arrow indicates the 15-minute interval ending at 1:00 PM on Thursday, the earliest starting time for insider trades in the sample.
Median Change v. Avg. Wednesday
1100% 900%
traded non-traded
700% 500% 300% 100% -100% Wednesday
↑ Thursday
Friday
Figure 2 Three-Day Median Price Changes This figure shows the median percentage price change from the open on the Wednesday preceding the release of the relevant IWS column until the close on the Friday when the magazine is publicly released. These data are measured in 15-minute intervals relative to the average price on Wednesday. This plot is drawn for the 21 stocks that at least one of the brokers traded (“traded”) and 44 that no insider traded according to the SEC complaint (“not traded”), using transactions data summarized in 15-minute intervals. Only stocks not mentioned in another news source on the insider trading day (Th) or the day before (W) are included. The two vertical lines represent the end of the first (W) and second (Th) trading days. The arrow indicates the 15-minute interval ending at 1:00 PM on Thursday, the earliest starting time for insider trades in the sample.
13%
Median Change v. Avg. Wednesday
11%
traded non-traded
9% 7% 5% 3% 1% -1%
Wednesday
Thursday ↑
Friday
Figure 3 Average Holding Period for "Inside Wall Street" Stocks Traded by Brokers The average holding period is computed for stocks traded by stockbrokers who had advance copies of the "Inside Wall Street" column in Business Week magazine. The holding period decreased significantly over the period of trading, reflecting learning on the part of these brokers about the temporary nature of the Business Week "bounce."
7.00 6.00
Days
5.00 4.00 6.73
3.00 4.67
2.00
3.17
1.00
1.67
0.00 June/July
Aug/Sept
Oct/Nov
Dec/Jan
