Private Information, Earnings Announcements and Trading Volume. Stock Chat on the Internet: A Public Debate About Private Information

Transcription

1 Private Information, Earnings Announcements and Trading Volume or Stock Chat on the Internet: A Public Debate About Private Information Peter D. Wysocki * University of Michigan Business School 701 Tappan Street Ann Arbor, MI Telephone: (734) Fax: (734) wysockip@umich.edu This version: April 2000 Abstract This paper empirically tests Kim and Verrecchia s (1997) predictions on the relation between trading volume during an earnings announcement and the amount of investor private information prior to and during the earnings announcement. I use pre-announcement and announcement period message-posting activity on The Motley Fool stock chat boards as a proxy for the levels of investor private information in the two periods. Consistent with the Kim and Verrecchia predictions, I find that announcementperiod trading volume is increasing in pre-announcement and announcement-period posting activity and the absolute value of price change during the announcement period. More interestingly, I find that the relation between announcement-period trading volume and absolute value of price change is decreasing in the amount of announcement-period private information. These results generally support Kim and Verrecchia s bifurcated model of private trading activity around anticipated earnings announcement events. * I would like to thank Ro Verrecchia for helpful discussions that lead to this investigation. Comments and suggestions by Scott Richardson, S.P. Kothari, and seminar participants at the Michigan Business School are greatly appreciated.

2 1. Introduction This paper investigates the empirical relation between trading volume during an earnings announcement and the level of investor private information prior to and during the earnings announcement. Kim and Verrecchia (1997) provide two basic predictions on trading volume responses to the two types of private information. First, if there exists exclusively pre-announcement private information, expected trading volume is predicted to have a linear relation to the absolute value of price change, with a zero intercept. Second, if there exists exclusively event-period private information, expected trading volume is predicted to be independent of the absolute value of price change. While the two predictions appear to be clear-cut, Kim and Verrecchia concede that finding empirical proxies for the two types of differential informedness is bedeviled by a variety of problems. I present a novel empirical test of the Kim and Verrecchia model using a new set of data based on the level of message-posting activity on stock chat boards on the Internet. I argue that unexpected daily changes in message-posting volume is a direct proxy for changes in the level of investor private information. This constantly-changing measure of investor private information allows for a simple, direct test of trading volume responses to both pre-announcement and announcement-period private information. Consistent with the predictions of Kim and Verrecchia, I find that announcement-period trading volume is positively associated with the proxies for (1) the level of pre-announcement private 2

3 information, and (2) the level of announcement-period private information. Similar to prior research, I also find that trading volume is positively associated with the absolute value of price change in the announcement period. Of greater interest, however, is my empirical finding that the relation between trading volume and absolute value of price change is (1) increasing in the level of pre-announcement private information, but (2) decreasing in the level of announcement private information. The finding is confirmed with follow-up empirical tests that use the relative level of private information in the preannouncement period and the announcement period. These combined results support Kim and Verrecchia s contention that event-period private information leads to a weaker association between expected trading volume and the absolute value of price change during an earnings announcement. My empirical model and the associated results have a number of implications. First, they shed additional light on the following empirical regularities: (1) trading volume is positively related to the absolute value of price change around an earnings announcement (see, for example, Atiase and Bamber, 1994), and (2) increases in trading volume are observed during announcements in the absence of price changes (see, for example, Kandel and Pearson, 1995). My results are supportive of Kim and Verrecchia s richer description of real markets in which multiple types of private information lead to trade during earnings announcements. Second, my proxy for private information based on Internet message-posting activity shows promise for future empirical studies requiring a up-to-date measure of the level of investor private information. Third, the empirical results provide additional information on the new phenomenon of Internet message postings about publicly-traded companies. Consistent with Wysocki (1998), I show that message posting-activity is associated with trading activity and appears to 3

4 reflect investor information rather than just random noise. The results in this study lead to a better understanding of the information content of message-posting activity on stock chat boards. The remainder of the paper is organized as follows. The next section discusses the Kim and Verrecchia (1997) model and develops the research hypotheses. Section 3 describes the proxies for investor private information and the specific empirical procedures used to test the hypotheses. Section 4 discusses the empirical results. A summary and conclusions are presented in section Theory and hypotheses This study s empirical analysis is motivated by the Kim and Verrecchia (1997) model, which is summarized as follows. Each investor is assumed to employ private information in anticipation of and in conjunction with an earnings announcement. The former is referred to as pre-announcement information and the latter is event-period information. The two types of information arise when (1) the anticipated event motivates investors to gather pre-announcement private information, and (2) eventperiod private information is used to interpret the earnings disclosure. In real markets, it is expected that both types of information will co-exist surrounding an earnings announcement event. The main objective of Kim and Verrecchia is to explain the relation between trading volume and event-period and pre-announcement private information. They characterize two polar cases to describe the trading implications of the two types of information. The first polar case assumes the absence of event-period private information. Under this scenario, the only thing that generates trade is the fact that 4

5 investors have differing quality of pre-announcement private information. Therefore, investors will react differently to the earnings announcement. This result is basically the idea behind Kim and Verrecchia (1991a) and implies that trading volume is related positively to both the absolute value of price change and the degree of differential informedness in the pre-announcement information period. Therefore, when event-period information is ignored, expected volume is predicted to be linear in price change with zero intercept. Atiase and Bamber (1994) provide empirical evidence consistent with this prediction. On the other hand, in examining the empirical validity of this relation, Kandel and Pearson (1995) document significant abnormal trading volume when the announcement period return is effectively zero. Consequently, Kandel and Pearson call into question the assumption that earnings announcements are interpreted commonly across all investors, as opposed to differentially. The Kandel and Pearson result gives rise to the other polar case described by Kim and Verrecchia. Under this scenario, there exists only event-period private information. Kim and Verrecchia show that with only event-period private information trading volume is not directly related to the absolute value of price change. One would not expect to directly observe either of the extreme polar cases of private information around an earnings announcement. The reality is the an anticipated earnings announcement would generate both types of private information and trading volume would arise form a combination of the polar cases described above. 5

6 Given the rich hypotheses presented in Kim and Verrecchia, the objective of this paper is to empirically test the hypotheses. The difficulty, however, is finding reasonable proxies for the level of investor private information in the pre-announcement period and the announcement period. As discussed by Kim and Verrecchia, this is not a trivial task. Moreover, previous studies on trading volume surrounding earnings announcements provide little guidance on this issue. For example, Utama and Cready (1997) use cross-sectional differences in institutional ownership as a measure of the level of cross-investor variation in private predislcosure information. Unfortunately, this cross-sectional proxy cannot differentiate between pre-announcement and announcement-period private information. The lack of satisfactory empirical proxies for the two types of private information lead me to a search for different solution. In the next section, I present and discuss a new measure for pre-announcement and announcement-period information. 3. Empirical Procedures 3.1 Proxies for private information in the pre-announcement and announcement periods Stock message boards, also known as investor chat rooms, are sites on the World Wide Web and other on-line services where individuals discuss publicly-traded companies by posting messages that can be read by anyone visiting the board. Message board posting activity has grown from almost nothing two years ago to thousands of messages each day in recent months (see, for example, Harmon, 1998, and Bennett, 1998). What is unclear is what this message-posting activity represents and whether it has any relation to or impact on financial markets. 6

7 A quick inspection of message content might lead one to conclude that many posting are pure noise. Some postings and subsequent replies often degenerate into trash talk and personal attacks with only limited connection to the stock. On the other hand, message postings can evolve into sophisticated debates about company financial disclosures. For example, minutes after a firm releases its earnings, posters often quickly begin to dissect, interpret, and debate the financial report on-line. Posting activity clearly reflects differences in opinion among investors about the meaning of information releases. For example, consider the following message post by an individual with the pseudonym on the Dell Computer message board on the Motley Fool Message Board Site: AUTHOR: LUNDEX SUBJECT: Dell PE DATE: 2/2/99 MESSAGE: Now that Dell s PE is up to 118, it s time to sell! This message received a flurry of replies including one from an individual going by the pseudonym RLIKES : AUTHOR: RLIKES SUBJECT: Re: Dell PE DATE: 2/2/99 MESSAGE: Did I miss something? What is so special about 118? Why not 117 or 119? <has mental picture of LUNDEX thinking still a good deal still fine up! SELL!>. I just wish I 7

8 could have invested in EBAY (the most disgustingly overvalued company on the face of the earth, seriously) when it had a PE of 118. Would have had a fun run to a PE of 10,000. Wysocki (1998) presents some of the first evidence on the determinants and impact of message-posting activity on the Internet. He finds that message-posting activity is highest for firms with extreme past returns and accounting performance, high price-earnings and market-to-book ratios, high volatility and trading volume, high analyst following and low institutional holdings. In addition, he finds that daily changes in posting activity predict next day stock trading volume and volatility. At a purely descriptive level, these results indicate that message-posting volume is related to underlying firm characteristics and market activity. The empirical findings also show that posting activity is not just noise, but is associated with stock market information flows. Wysocki shows that earnings announcement events and trading volume are associated with increased positing activity. These findings suggest that posting activity is related to the arrival of information in the market and the interpretation of this information by market participants (see, for example, Kim and Verrecchia, 1991a and 1991b; Harris and Raviv, 1993; Kandel and Pearson, 1995) Given the empirical regularities described in Wysocki (1998), I argue that daily level of message-posting activity for a particular firm would be a suitable proxy to capture the level of investor private information about the firm. For example, posting volume is highest for high technology firms, firms with high price-earnings and market to book ratios and around earnings announcements. This is consistent with investors focussing on firms with the largest information asymmetry and the greatest 8

9 likelihood of future information flows. Many of these same characteristics also determine analyst following and activity (see, for example, Bhushan, 1989). Not surprisingly, Wysocki finds that analyst following is positively associated with posting activity. Therefore, message-posting activity can be viewed as a rational phenomenon where investors use Internet message boards to gather, process, and interpret private information. The nice feature of message posting activity is that it is changes on a daily basis and can be tracked for a broad range of publicly-traded companies. As of April 15, 1999, The Motley Fool Web Site ( had 2,995 active stock message boards. The message boards contain stocks traded on the NYSE, AMEX and NASDAQ stock exchanges. Figure 1 provides an example of a series of message posts for Dell Computer, one of the most active stock message boards. This particular series of posts is around the release of Dell s quarterly earnings information on November 12, The summary message board identifies the time, date, author, and topic of each message and whether the message is on a new topic or in reply to previous message. In the following empirical analysis, I restrict the sample to those firms with the highest cumulative posting activity. Wysocki (1998) finds that message posting activity is highest for firms with extreme past returns and accounting performance, high price-earnings and market-to-book ratios, high volatility and trading volume, high analyst following and low institutional holdings. Therefore, these cases appear to be good candidates for firms that are likely to exhibit large private information effects and have significant variation in this private information over time. I propose to use the level of abnormal daily message-posting activity as a measure of the daily level of investor private information. This measure will 9

10 be used to characterize private information in both the pre-announcement and announcement periods around an anticipated earnings release. The validity of my inferences in this study is subject to the caveat that the empirical proxy for the level of private information is a reasonable surrogate for the theoretical construct. For example, message postings activity has the potential to proxy for other information such as investor sentiment, noise, or level of disagreement that is unrelated to private information. Therefore, the results in this paper should be interpreted with the caveat that the proxy could measure some other factor that is correlated with announcement-period trading volume and is not capturing the effects of private information. 3.2 Sample I examine quarterly earnings announcements for a set of 55 firms made between April 15, 1998 and April 15, The 55 firms were identified using the following selection criteria: (1) the firm was among the 55 publicly-traded firms with the highest number of cumulative message postings on the Motley Fool stock message board site as of April 15, 1999, (2) complete historical daily trading data is available for the firm from the Dow Jones Interactive Web Site between April 15, 1998 and April 15, 1999, (3) quarterly earnings announcement dates were available for the firm from the Bloomberg Database for the full sample period. Table 1 lists the 55 firms with the highest number of cumulative message postings on the Motley Fool stock message board site as of April 15, The 5 firms dropped from the final sample, and the reason for their exclusion, are identified in the table. The final sample contains 197 quarterly earnings-announcement events for the sample of 50 firms. 10

11 3.3 Empirical model Based on the preceding analysis and prior empirical models of trading response (see, for example, Atiase and Bamber, 1994; Utama and Cready, 1997), I estimate the following linear model using OLS regression: VOL(ANN) i = α + β 1 ARET(ANN) i + β 2 MESS(PRE) i + β 3 MESS(ANN) i + β 4 MESS(PRE) i *ARET(ANN) i + β 5 MESS(ANN) i *ARET(ANN) i (1) where i VOL(ANN) designates a firm-announcement; is the ratio of average daily trading volume in the 5-day announcement period to the median daily trading volume in the 20 trading days prior to the 5-day pre-announcement period; ARET(ANN) is the absolute value of the cumulative return in the 5-day announcement period; MESS(PRE) is the ratio of average daily messages posted in the 5-day pre-announcement period to the median messages posted in the 20 trading days prior to the pre-announcement period; MESS(ANN) is the ratio of the average daily messages posted in the 5-day announcement period to the median messages posted in the 20 trading days prior to the pre-announcement period. as follows: A time-line of the non-announcement, pre-announcement period, and announcement periods is Time-Line of Non-announcement, Pre-announcement and Announcement Periods 11

12 (Day t=0 defines earnings announcement day) Non-announcement Period Pre-announcement Period Announcement Period 20 trading-day window t=-5 t=-4 t=-3 t=-2 t=-1 t=0 t=1 t=2 t=3 t=4 The predictions are that trading volume is increasing in both pre-announcement and eventperiod private information. Hence, β 2 and β 3 are hypothesized to be positive. A positive value on β 2 is consistent with both Kim and Verrecchia (1991a), as well as the follow-up model in Kim and Verrecchia (1997). Positive coefficients would support their theoretical proposition that trading volume response to a public announcement is an increasing function of private pre-announcement and eventperiod information. Consistent with Atiase and Bamber (1994), ARET is included as a control for the average change in investors beliefs, and, as is consistent with Kim and Verrecchia (1991a), a positive value for β 1 is predicted. Based on Kim and Verrecchia s (1997) model of event-period private information, I examine the relation between trading volume and the absolute value of returns as a function of announcement-period message postings. Therefore, the coefficient on the second interactive variable, β 5, is predicted to be negative. In other words, the relation between trading volume and the absolute value of returns is expected to be decreasing in the level of announcement-period private information. As a robustness check, I also estimate the following linear regression specification: VOL(ANN) i = α + β 1 ARET(ANN) i + β 2 [MESS(ANN) i /MESS(PRE) i ] + β 3 [MESS(ANN) i /MESS(PRE) i ]*ARET(ANN) i (2) 12

13 where the variables are defined as in equation (1). In this case, I measure the relative amount of preannouncement vs. event-period private information 1. In this case, β 3 is predicted to be negative because the relation between trading volume and the absolute value of returns is expected to be decreasing in the relative amount of event-period private information. 3.4 Definition of non-announcement, pre-announcement, and announcement periods Prior research suggests that announcement days have abnormally high trading activity compared to other trading days in the year. This research also indicates that the abnormal trading volume persists up to five days after the earnings announcement (Morse, 1981; Bamber, 1987). I define the announcement-period as a five trading-day window starting on the earnings announcement date (t=0, +4). For symmetry, I define the pre-announcement period as the five trading-day window prior to the announcement (t=-5,-1), where t=0 denotes the announcement date as reported on the Bloomberg database. The use of a five trading-day window for the announcement and pre-announcement periods also removes any day-of-the-week trading effects. The non-announcement period is defined as the twenty trading-day window prior to the preannouncement period. This extended window is used as a benchmark for trading and message-posting activity outside of the announcement periods. 1 It can be argued that the predictions based on Kim and Verrecchia s (1997) model of pre-announcement and event period private information are best characterized as relative comparisons of the level of the two types of information. 13

14 As a robustness check, I also replicate the empirical analysis using (1) pre-announcement and announcement windows from 3 to 7 days in length, and (2) announcement windows beginning the day after the listed earnings announcement date (t = +1). The empirical results (not reported) for these other specifications are qualitatively similar to the results presented in following sections. 3.5 Measurement of variables Trading volume response Earnings announcement trading volume is number of shares of firm i s shares traded on day t, averaged over the five-day announcement period (t = 0,4). I use the median daily trading volume in the 20-day non-announcement period as a measure of expected trading volume for firm i. As in Bamber and Cheon (1995), I estimate the relative unexpected volume in the announcement period as the ratio of the announcement trading volume to the expected trading volume, which is the median non-announcement-period trading volume. This metric is denoted VOL(ANN). I estimate the unexpected volume in the pre-announcement period as the ratio of the pre-announcement trading volume to the expected trading volume. This metric is denoted VOL(PRE). If there is no trading volume reaction around earnings announcements then these ratios should be close to one, while if there is more trading volume during the announcement they will exceed one Abnormal returns 14

15 Abnormal returns during the announcement period are cumulated over the five-day announcement window. For each day t during the announcement window (t=0 to 4), a firm s abnormal return is computed using the market model. I use an equally-weighted index of the daily returns for the 50 stocks in the sample as a proxy for the return on the market. I then estimate the market-model parameters over each earnings announcement period for each firm 2. Price responses to earnings announcements are measured using the absolute value of the cumulated abnormal returns over the fiveday announcement window. This metric is denoted ARET(ANN). Abnormal returns are calculated in a similar manner for the five-day pre-announcement period. The market responses in the pre-announcement windows are measured using the absolute value of the cumulated abnormal returns over the five-day pre-announcement window. This metric is denoted ARET(PRE) Message posting activity Announcement-period message posting volume is measured the average number of messages posted in the five-day (t=0 to 4) earnings-announcement window. I estimate the relative unexpected message volume in the announcement period as the ratio of the announcement-period message volume to the expected volume, which is the median daily message volume in the 20-day non-announcement window. This metric is defined as MESS(ANN). If there is no increase in message posting activity 2 Over the 1 year sample period, most of the 50 firms had 4 different earnings announcement dates listed on the Bloomberg database. 15

16 during the earnings-announcement period the ratio should be close to one, while if there is more message-posting activity during the announcement-period it should exceed one. Pre-announcement message posting volume is measured the average number of messages posted in the five-days (t=-5 to -1) prior to an earnings announcement event. I estimate the relative unexpected message volume in the pre-announcement period as the ratio of the per-announcement message volume to the expected volume, which is the median non-announcement period daily message volume. This metric is defined as MESS(PRE). Again, if there is no increase in message posting activity prior to an earnings announcement the ratio should be close to one. 4. Results 4.1 Descriptive statistics Table 2 presents descriptive statistics for the sample. As shown in Panel B, the unexpected trading volume metrics, VOL(ANN) and VOL(PRE), indicate that pre-announcement and announcement-period daily trading volume exceeds median daily trading volume in the nonannouncement period by more than 110% and 70%, respectively. This increase is consistent with previous research on trading activity surrounding earnings announcement events. The unexpected message posting metrics, MESS(ANN) and MESS(PRE), indicate that announcement and pre-announcement period daily message volume on The Motley Fool boards exceeds median daily message volume in the non-announcement period by a factor of five and a factor 16

17 of three, respectively. This is consistent with the evidence in Wysocki (1998) that shows a substantial increase in message-posting activity surrounding earnings-announcement events. The average value of the absolute value of cumulative abnormal return in the 5-day announcment window is 5.5%. This variable is significantly positively skewed. In fact, the maximum abnormal return is 58%. Clearly, major news releases for a number of firms occurred during their earnings announcement periods. The correlations among the variables are presented in Panel B of Table 2. All of the variables are positively correlated with each other. The positive correlation between trading volume, the absolute value of cumulative stock returns, and message posting volume are consistent with the predicted relation between the variables. However, multiple regression analysis is required to determine the incremental associations between trading volume, absolute value of returns and message posting activity. The correlation table indicates that pre-announcement and announcement period trading volume and absolute value of cumulative stock returns are correlated. This indicates that volume and volatility is persistent around earnings-announcement events. This could be a confounding effect for the regression outlined in equation (1). I control for these variables in a robustness check described in section Regression results 17

18 Table 3 presents the regression for the primary empirical test. The explanatory variables outlined in equation (1) explain approximately 18% of the variation in unexpected trading volume in the announcement period. Consistent with Kim and Verrecchia (1997) hypotheses, the coefficients on β 2 and β 3 are positive and significant. This is consistent with pre-announcement and announcement-period private information leading to announcement period trading volume. In particular, a 100% increase in unexpected pre-announcement posting activity is associated with a 46% increase in announcementperiod trading volume. A 100% increase in unexpected announcement-period posting activity is associated with a 62% increase in announcement-period. The coefficient on ARET(ANN) is positive and significant. This is consistent with the prediction that trading volume is linearly related to absolute value of price change and consistent with prior empirical work. The most interesting empirical results relate to the interactive terms in the regression. The interactive terms indicate that the relation between trading volume and absolute value of price change is (1) increasing in the level of pre-announcement private information (although it is not significant), but (2) decreasing in the level of announcement private information. This result supports Kim and Verrecchia s contention that event-period private information leads to a weaker association between expected trading volume and the absolute value of price change during an earnings announcement. 4.3 Robustness and specification checks 18

19 I examine the sensitivity of the results to the inclusion of pre-announcement trading volume and the absolute value of pre-announcement cumulative returns. I perform this check to rule out the possibility that pre-announcement message posting volume is just proxying for market activity in the preannouncement period which persists into the announcement period. Table 4 replicates the previous regression results using the same model but also including pre-announcement period trading volume and absolute value of pre-announcement cumulative returns as additional explanatory variables. The coefficients on VOL(PRE) and ARET(PRE) are positive and significant. This indicates that trading volume and return volatility is persistent. However, the sign and significance of the coefficients in the original set of explanatory variables remains unchanged. Therefore, the results in the original regression do not appear to be spurious. The predictions of Kim and Verrecchia examine the relative amount of private information in the pre-announcement and announcement periods. As discussed in section 2, an alternate empirical specification would be the relative amount of message posting activity in the pre-announcement and announcement periods. Therefore, I estimate the data using model (2). The regression results are presented in Table 5. Similar to the previous results, I find that the announcement period trading volume is increasing in the absolute value of announcement period returns. I also find that announcement period trading volume is increasing in the ratio of announcement period to pre-announcement period message postings. A 100% increase in the amount of posting activity in the announcement period (relative to the pre-announcement period) leads to a 122% increase in unexpected posting activity. Therefore, it appears that higher relative private information in the announcement period leads to greater trading volume in the announcement period. Alternatively, greater levels of disagreement about information in 19

20 earnings announcements appears to dominate in explaining trading volume in the announcement period. Finally, the positive relation between trading volume and absolute value of returns is decreasing in the relative amount of announcement-period private information. This result is consistent with the previous regression and with the prediction of Kim and Verrecchia (1997). 5. Conclusion and discussion This study provides a direct empirical test of Kim and Verrecchia s (1997) model of trading volume responses to pre-announcement and event-period private information. Kim and Verrecchia (1997) provide two simple predictions on trading volume responses to the two types of private information. However, direct tests of the predictions would prove difficult using existing empirical proxies for investor private information because the proxies cannot easily differentiate between preannouncement and event-period private information. I use a novel empirical proxy to separately identify investor private information preceding and during an earnings announcement event. This proxy is based on the level of message-posting activity on stock chat boards on the Internet. I use unexpected daily changes in message-posting volume on stock chat boards as a direct proxy for changes in the level of investor private information. Consistent with the predictions of Kim and Verrecchia, I find that announcement-period trading volume is positively associated with the proxies for (1) the level of pre-announcement private information, and (2) the level of announcement private information. Similar to prior research, I also find that trading volume is positively associated with the absolute value of price change in the announcement period. Of greater 20

21 interest, however, is my empirical finding that the relation between trading volume and absolute value of price change is (1) positively related to the level of pre-announcement private information, but (2) negatively related to the level of announcement private information. These results support Kim and Verrecchia s contention that event-period private information leads to a weaker association between expected trading volume and the absolute value of price change during an earnings announcement. My results extend the prior empirical literature on trading volume and directly differentiate between pre-announcement and event period private information. The findings also suggest that the new phenomenon of message-posting activity can be a useful source of data on investor beliefs and private information about publicly-traded companies. In future drafts of this paper, I plan to expand the data analysis to a larger sample of firms when the newest editions of CRSP and COMPUSTAT tapes are issued. This will allow for easier tabulation of stock trading activity and earnings announcement dates for an expanded list of firms and dates. 21

22 References Atiase, R. and L. Bamber, 1994, Trading volume reactions to annual accounting earnings announcements: The incremental role of predisclosure information asymmetry, Journal of Accounting and Economics 17, Bamber, L. 1987, Unexpected earnings, firm size and trading volume around quarterly earnings announcements, The Accounting Review 62, Bamber, L., and S. Cheon, 1995, Differential price and volume reactions to accounting earnings announcements, The Accounting Review 70, Bennett, J., 1998, Traffic on financial web pages rises when the market falls, Dow Jones News Service, September 1, Goldstein, A., 1998, Money messages: Electronic message boards are a good way to get investing facts and fiction, The Dallas Morning News, August 3, 1998, 1D. Harmon, A., 1998, The market turmoil: Investors on line, The New York Times, September 1, 1998, 6. Harris, M. and A. Raviv, 1993, Differences of opinion make a horse race, Review of Financial Studies 6, Kandel, E. and N. Pearson, 1995, Differential interpretation of public signals and trade in speculative markets, Journal of Political Economy 103, Kim, O. and R. Verrecchia, 1991a, Trading volume and price reactions to public announcements, Journal of Accounting Research 29, Kim, O. and R. Verrecchia, 1991b, Market reaction to anticipated announcements, Journal of Financial Economics 30, Kim, O. and R. Verrecchia, 1997, Pre-announcement and event-period information, Journal of Accounting and Economics 24, Morse, D., 1981, Price and trading volume reaction surrounding earnings announcements: a close examination, Journal of Accounting Research 19, Utama, S. and W. Cready, 1997, Institutional ownership, differential predisclosure precision and trading volume at announcement dates, Journal of Accounting and Economics 24,

23 White, H., 1980, A heteroscedasticity-consistent covariance matrix estimator and a direct test for heteroscedasticity, Econometrica 48, Wysocki, P., 1998, Cheap talk on the web: The determinants of postings on stock message boards, University of Michigan Business School Working Paper. 23

24 Table 1 List of Sample Firms Used in Empirical Tests Top 55 stocks with active message boards on the Motley Fool Investor Web Site ( As of April 15, 1999, there were 2995 stocks with active message boards on Motley Fool Site. Rank Company Name Ticker Cumulative Posts 1 Dell Computer Corporation DELL Iomega Corporation IOM Amazon.com, Inc. AMZN America Online Inc. AOL CMGI, Inc. CMGI Microsoft Corp. MSFT At Home Corporation ATHM Tag Heuer Intl THW 7242 Dropped - no earnings data 9 Coca-Cola Company, The KO Berkshire Hathaway BRK.A 6185 Dropped - no trading data 11 Intel Corporation INTC Yahoo! Inc. YHOO Cisco Systems, Inc. CSCO Compaq Computer Corp. CPQ Apple Computer, Inc. AAPL Philip Morris Companies MO E*TRADE Group, Inc. EGRP Starbucks Corporation SBUX Pfizer, Inc. PFE RealNetworks, Inc. RNWK Lucent Technologies Inc. LU Disney DIS Ebay Inc. EBAY 2307 Dropped - incomplete trading 24 Oracle Corporation ORCL EMC Corporation EMC AT&T Corporation T Qwest Communications QWST Lycos, Inc. LCOS Home Depot, Inc., The HD Netscape Communications NSCP Ford Motor Company F Knight/Trimark Group, Inc. NITE 1461 Dropped - incomplete trading 33 Harley-Davidson, Inc. HDI Sun Microsystems, Inc. SUNW Egghead.com EGGS Gateway 2000, Inc. GTW Rambus, Inc. RMBS Boeing Company BA Infoseek Corporation SEEK Excite, Inc. XCIT PeopleSoft, Incorporated PSFT Advanced Micro Devices AMD Borders Group, Inc. BGP Wal-Mart Stores, Inc. WMT Aware, Inc. AWRE DoubleClick Inc. DCLK AmeriTrade Holding Corp. AMTD Safeskin Corporation SFSK IBM IBM Rainforest Cafe, Inc. RAIN PepsiCo, Inc. PEP WorldCom, Inc. WCOM Echelon Corporation ELON 952 Dropped - incomplete trading 54 Cendant CD Gap, Inc., The GPS

25 Table 1 - continued The sample consists of 50 firms with (1) the highest cumulative message posting volume and on the The Motley Fool stock message boards as of 4/15/99, (2) complete daily stock trading data from Dows Jones Interactive Web Site, (3) quarterly earnings announcement dates for the period 4/15/98-4/15/99 on the Bloomberg database. The final sample consists of 197 earnings announcement events. 25

26 Table 2 Descriptive Statistics for the Sample The sample consists of 197 quarterly earnings announcement events for 50 publicly-traded firms between 4/15/98-5/15/99. Panel A - Variable Values Mean Median Std. Dev. Minimum Maximum VOL(ANN) VOL(PRE) ARET(ANN) ARET(PRE) MESS(ANN) MESS(PRE) Panel B - Correlations VOL(ANN) VOL(PRE) ARET(ANN) ARET(PRE) MESS(ANN) MESS(PRE) VOL(ANN) 1 VOL(PRE) ARET(ANN) ARET(PRE) MESS(ANN) MESS(PRE) VOL(ANN) is the ratio of average daily trading volume in the five-day announcement period to the median daily trading volume in the 20 trading days prior to the five-day pre-announcement period. VOL(PRE) is the ratio of average daily trading volume in the five-day pre-announcement period to the median daily trading volume in the 20 trading days prior to the five-day pre-announcement period. ARET(ANN) is the absolute value of the cumulative return in the five-day announcement period. ARET(PRE) is the absolute value of the cumulative return in the five-day pre-announcement period. MESS(PRE) is the ratio of average daily messages posted in the five-day preannouncement period to the median messages posted in the 20 trading days prior to the pre-announcement period. MESS(ANN) is the ratio of the average daily messages posted in the five-day announcement period to the median messages posted in the 20 days prior to the pre-announcement period. 26

27 Table 3 OLS Regression Results The sample consists of 197 quarterly earnings announcement events for 50 publicly-traded firms between 4/15/98-5/15/99. Regression model: VOL(ANN) i = α + β 1 ARET(ANN) i + β 2 MESS(ANN) i + β 3 MESS(PRE) i + β 4 MESS(ANN) i *ARET(ANN) i + β 5 MESS(PRE) i *ARET(ANN) i Variable Expected direction Coefficient Estimate White t-value Intercept? ARET(ANN) MESS(ANN) MESS(PRE) MESS(ANN)*ARET(ANN) MESS(PRE)*ARET(ANN) Adj. R % Number of Observations 197 VOL(ANN) is the ratio of average daily trading volume in the five-day announcement period to the median daily trading volume in the 20 trading days prior to the five-day pre-announcement period. ARET(ANN) is the absolute value of the cumulative return in the five-day announcement period. MESS(PRE) is the ratio of average daily messages posted in the five-day pre-announcement period to the median messages posted in the 20 trading days prior to the pre-announcement period. MESS(ANN) is the ratio of the average daily messages posted in the five-day announcement period to the median messages posted in the 20 days prior to the pre-announcement period. 27

28 Table 4 OLS Regression Results - Robustness Check The sample consists of 197 quarterly earnings announcement events for 50 publicly-traded firms between 4/15/98-5/15/99. Regression model: VOL(ANN) i = α + β 1 ARET(ANN) i + β 2 MESS(ANN) i + β 3 MESS(PRE) i + β 4 MESS(ANN) i *ARET(ANN) i + β 5 MESS(PRE) i *ARET(ANN) i + β 6 VOL(PRE) i + β 7 ARET(PRE) i Variable Expected direction Coefficient Estimate White t-value Intercept? ARET(ANN) MESS(ANN) MESS(PRE) MESS(ANN)*ARET(ANN) MESS(PRE)*ARET(ANN) VOL(PRE) ARET(PRE) Adj. R % Number of Observations 197 VOL(ANN) is the ratio of average daily trading volume in the five-day announcement period to the median daily trading volume in the 20 trading days prior to the five-day pre-announcement period. VOL(PRE) is the ratio of average daily trading volume in the five-day pre-announcement period to the median daily trading volume in the 20 trading days prior to the five-day pre-announcement period. ARET(ANN) is the absolute value of the cumulative return in the five-day announcement period. ARET(PRE) is the absolute value of the cumulative return in the five-day pre-announcement period. MESS(PRE) is the ratio of average daily messages posted in the five-day preannouncement period to the median messages posted in the 20 trading days prior to the pre-announcement period. MESS(ANN) is the ratio of the average daily messages posted in the five-day announcement period to the median messages posted in the 20 days prior to the pre-announcement period. 28

29 Table 5 OLS Regression Results - Alternate Specification The sample consists of 197 quarterly earnings announcement events for 50 publicly-traded firms between 4/15/98-5/15/99. Regression model using relative message-posting volume: VOL(ANN) i = α + β 1 ARET(ANN) i + β 2 [MESS(ANN) i /MESS(PRE) i ]+ β 3 [MESS(ANN) i /MESS(PRE) i ]*ARET(ANN) i Variable Expected direction Coefficient Estimate White t-value Intercept? ARET(ANN) MESS(ANN)/MESS(PRE) [MESS(ANN)/MESS(PRE)] *ARET(ANN) Adj. R % Number of Observations 197 VOL(ANN) is the ratio of average daily trading volume in the five-day announcement period to the median daily trading volume in the 20 trading days prior to the five-day pre-announcement period. ARET(ANN) is the absolute value of the cumulative return in the five-day announcement period. MESS(PRE) is the ratio of average daily messages posted in the five-day pre-announcement period to the median messages posted in the 20 trading days prior to the pre-announcement period. MESS(ANN) is the ratio of the average daily messages posted in the five-day announcement period to the median messages posted in the 20 days prior to the pre-announcement period. 29