The Internet is clearly playing an ever-increasing role in financial markets and personal finance. The six largest Internet brokerages cumulatively boasted over 12 million accounts in 1999 and grew significantly in 2000. And investors now benefit from a wide assortment of financial information available online, ranging from Securities and Exchange Commission documents to financial sites like the Motley Fool. • One of the more interesting phenomena on the web has been the growth of stock-related chat rooms and bulletin boards, which facilitate discussion among thousands of investors. Bulletin boards, which are not live forums, allow users to post messages for retrieval by others at a later time. A typical site contains distinct bulletin boards for each market security that users can discuss.

 

ecently, the press has sensationalized the activity in these forums, linking it to egregious examples of stock-price manipulation. For example, in February 1999, the stock price of a small Milwaukee-based toy company, Alottafun Inc., soared 382% based on speculation started in Internet chat rooms. Despite such examples, the vast majority of the discussion involves investors honestly expressing their opinions on securities markets.

Over the past several years, analysts and academics have tried to figure out a means to place appropriate values on Internet stocks. And some have focused on whether message-board activity has any bearing on stock prices. So, I thought it would be useful to examine the relationship among the volume, quantity, and quality of the opinions expressed on message boards about individual stocks, and the movement in stock prices. In other words, I sought to determine whether message board activity helps predict stock returns and/or trading volume.

I chose to focus on a single site – RagingBull.com – in part because it is extremely popular. Between April and November 1999, the site’s membership tripled in size to 300,000, while averaging six million daily page views. In addition, RagingBull.com has some unique attributes. The site has categories for different ticker symbols. And it includes an “optional disclosure” feature, which lets users clearly indicate their opinion on the short and long-term prospects of the stock by selecting long, short, or no position. Similarly, users can issue strong buy, buy, hold, sell, and strong sell ratings for both the short-term and the long-term.

Table 1: Click for larger image

Because much of the discussion on sites like RagingBull.com revolves around high-technology and Internet companies, I chose to examine the postings and activity in a group of 73 Internet service companies drawn from Zacks’ Internet Services sector group. This group included well-known, large firms like Yahoo! (market capitalization $114.8 billion at the time) and many obscure companies, like Biznessonline.com (worth $53.1 million at the time). The sample had a medium market capitalization of $1.12 billion.

For this study, I downloaded some 181,633 messages posted between April 17, 1999, the day when the opinion-disclosure feature was added to RagingBull, and February 18, 2000. Of that total, 43,794 (24.1%) contained short-term opinions, 37,810 (20.8%) had long-term opinions, and 52,812 (29.1%) included a general “voluntary disclosure.” Most stocks did not have a huge number of postings each day. The mean stock message board had an average of 7.6 posts daily, while the median message board had 2.5 messages per day. The maximum average number of daily postings was 103.6 – for CMGI Inc.

The next task was to calculate the average short-term opinion. Messages with short-term strong-buy recommendations were assigned a value of +2. Similarly, messages with short-term buy, hold, sell, and strong-sell recommendations were assigned values of +1, 0, -1, and –2, respectively. These opinion values were averaged on a daily basis to calculate the daily average opinion for each stock. The mean daily average opinion was 1.56, while the median was 1.64 – somewhere between a buy and strong buy.

I also weighted opinions for each stock on a daily basis. Each message with a short-term opinion was assigned a value according to the aforementioned scale. These opinions were added to calculate the daily weighted opinion, which was then averaged for each stock. The mean average daily weighted opinion was 6.09, while the median was 3.44. The standard deviation of the average daily weighted opinion value was 9.49. The maximum was 56.64 (CMGI Inc.) while the minimum was 1.14 (TheGlobe.com).

Figures 3, 4, 5: Click for larger image

ext, I calculated the arithmetic average and standard deviation of daily returns for each stock during the sample period. The mean arithmetic average of daily return for the stocks was 0.677% and the median was 0.648%. The maximum average daily return was 2.53% (Be Free Inc.) and the minimum was -0.58% (Flashnet Communications). The average standard deviation of daily returns was 7.59% and the median was 7.39%. The maximum standard deviation was 13.37% (Cobalt Group) and the minimum was 4.80% (Cybercash). As might be expected, given the volatility of the Internet sector at the time, the average return and standard deviation of returns are very high compared to average values in the stock market during the sample period. The final piece of data was compiling the average trading volume for each stock.

I then subjected this data to two methodologies: a vector autoregression (VAR) analysis and an event study.

 

VAR Analysis

I also performed a vector autoregression (VAR) analysis – on a stock-by-stock basis – to examine the general relationship among stock returns, trading volume, message postings, and weighted opinion. This analysis showed that none of these factors was useful in predicting stock returns one day into the future. The analysis did show, however, that high trading volume days tended to precede days of high trading volume – and that low trading volume days tended to precede days of low trading volume. In other words, the opinion represented in bulletin-board messages were not helpful in predicting daily stock returns. This is consistent with market efficiency.

And days with high trading volume and positive weighted opinions are followed by days with greater message activity. Finally, weighted opinion is dependent on the number of messages and opinions posted on the previous day. Positive opinion days tend to follow days with positive opinions. The dependence of weighted opinion on the number of messages posted is consistent with the simple summation method used to calculate weighted opinion and the observation that each message board had positive average daily weighted opinions.

 

The Event Study

Since it was clear that bulletin-board opinions had no effect on stock prices in general, an event study was conducted. The event study attempted to answer the question: Does an unusual level of bulletin-board discussion measurably impact stock prices? Days with unusual levels of discussion were termed “event days” and were defined as those with message postings that exceeded the previous five-day average by at least two five-day standard deviations. (Event days in which fewer than 10 messages were posted were excluded from the sample.)

Over the past several years, analysts and academics have tried to figure out a means of placing appropriate values on internet stocks. And some have focused on whether message-board activity has any bearing on stock prices.

I examined two opinion metrics to determine the strength of opinion changes on the event day. The raw change in weighted opinion was calculated as the difference between the event-day weighted opinion and the average weighted opinion over the previous five days. The adjusted change in weighted opinion was calculated as the raw change in weighted opinion divided by the standard deviation of weighted opinion over the previous five days.

The event study found a total of 293 event days. Forty-seven of these days had opinions lower than the previous five-day average, and were classified as “negatives,” while 241 of the event day opinions were greater than the previous five-day opinion average, and were dubbed “positives.” (Five event days had opinion equal to the previous five-day average). These positives were further split in half. The “strong positives” category contained those with event days with the strongest opinion change. The “weak positives” contained the remaining event days – those with the weakest positive opinion change. Table 1 presents descriptive statistics for the changes in opinion on the event days.

 

Adjusted Returns and Abnormal Trading Volume

Because of the high and volatile returns in this sector, it was necessary to adjust the daily returns for industry returns. So, I adjusted the returns on my chosen 73-stock portfolio using the Philadelphia Stock Exchange (PSE) Internet Index. The industry adjusted return was defined as a stock’s daily return less the return on the PSE Internet Index.

Abnormal trading volume, which is defined as the percentage change in trading volume on a given day compared to the average trading volume, was computed for each ticker and each day during the sample period. A 20-trading-day period preceding the day in question was used to calculate the average trading volume.

 

The Results

So what did I find? Figures 2, 3, and 4 show the industry-adjusted returns and abnormal volume for a five-day period surrounding the event day. It is apparent that only strong-positive-opinion events classified using the raw change in weighted opinion show a statistically significant positive drift up to the event day (Figure 2). Returns for weak-positive-opinion events are statistically flat leading up to the event day. Negative-opinion event days seem to show a downward drift up to the event day, but the phenomenon is not statistically significant. On the event day, both strong and weak positives have statistically significant, positive industry-adjusted returns (Figures 2 and 3). Negative-opinion event days have a slightly negative industry-adjusted return, which is not statistically significant. Returns for all the opinion groups are statistically flat after the event day.

The results show that message board activity is definitely linked to stock price movements. However, abnormal message board activity does not help predict future stock price movements over a one-day or five-day window in the future.

Similarly, trading volume is normal leading up to the event day. But on and one day after the event day, there is a sharp increase in trading volume (Figure 4). The strongly positive raw change in weighted-opinion group shows the most significant increase in trading volume. For that group, approximately 160% more shares are exchanged on the event day than on the previous 20 days. Trading volume retreats to more normal levels approximately two days past the event day.

he results show that message board activity is definitely linked to stock price movements. However, abnormal message board activity does not help predict future stock price movements over a one-day or five-day window in the future. This observation is consistent with market efficiency. On the event day, strong-positive and weak-positive event days showed statistically significant returns in excess of the industry index. Therefore, abnormal message-board activity is coincident with abnormal stock returns. Using this methodology, however, it is impossible to determine whether activity on the message boards causes or is the result of abnormal returns on the stock.

 

Conclusions

So, can one predict future stock returns and performance based solely on message board activity? Not really. The event study shows that returns following abnormal Internet message-board activity are statistically insignificant. However, statistically significant positive returns precede the days with strong positive opinions and abnormal message board activity. Furthermore, stock returns and message-board opinions on days of abnormal message-board activity appear to be related.

These results are significant because they counter the conventional wisdom that Internet service stocks are valued irrationally. In general, message-board activity and opinion do not appear to impact stock prices in a significant, industry-adjusted fashion. Furthermore, abnormal message-board activity does not appear to predict significant abnormal returns. In sum, at least in the period I studied, the valuation of Internet service stocks appeared reasonable and consistent with market efficiency.

 

Robert Tumarkin, a 2000 NYU Stern graduate, is an analyst at Mezzacappa Management, a New York-based hedge fund.

The author wishes to acknowledge the assistance of Professor Robert Whitelaw.

The research discussed in this article was completed with a grant from the I. Glucksman Institute for Research in Securities Markets at Stern. The paper in its entirety will be published in a forthcoming issue of the Financial Analysts Journal.