eInformation: a clinical study of investor discussion and sentiment

Financial Management (Financial Management Association), Autumn, 2005 by Sanjiv Das, Asis Martinez-Jerez, Peter Tufano

Figure 1 shows the postings that followed these 16 events. Panel A shows that in the few hours immediately after a news event, posting volume rises, but then tails off over time. Panel B displays the composition of the posts over the first eight post-news hours.

[FIGURE 1 OMITTED]

We measure the nature of the discussion by categorizing each subsequent post into one of five possible categories (asks question, offers alleged fact, shares opinion, comments unrelated to news event, and spam/garbage). The first three categories are on-point postings, i.e., ones that relate to the news at hand. We see that for the first four hours after a news event, over two-thirds of all posts are on-point, and even eight hours later, about half are still discussing the news (as opposed to other issues or spam).

D. The On-Line Discussion Is a Mix of Questions, Answers, and Opinions

We categorize on-point posts as asking a question, offering an alleged fact, or proposing an opinion about the meaning of the news. For the first hour, we see more of a question-and-answer pattern, with a quarter of all posts and a third of the on-point posts either asking a question or supplying a fact. Over time, the discussion tends toward more analysis, i.e., interpreting facts, an observation that is consistent with our understanding of the primary function of the board and with our discussions with Glenn, the prolific Amazon poster.

The research design in this first experiment is biased in that it is conditioned on validated news, i.e., all of the stories we studied were real.

In the second analysis, we note and track rumors on the boards. In particular, we search our sample postings for messages related to mergers and acquisitions, which are material corporate events. The keywords we use for the search are merge, merger, hostile, acquisition, acquire, takeover, target, tender, offer, and stock swap. Our goal is to identify events with enough potential materiality that they might be ultimately reported.

We also searched for news related to the posting rumors in Factiva for the period January 1998 to August 1999 (i.e. six months before and after our sample period). In total, we identified 54 merger and acquisition rumors on the discussion boards, with seven of these meriting at least five posts. However, the rumors are almost entirely ungrounded in fact. In two instances, the press story announcing the rumor cited the Internet as the source of the rumors.

Just one of the 54 rumors preceded an announcement by the company of an actual merger. Nine others preceded similar rumors in the business press, but not an actual transaction. One preceded a press denial by the company. The majority (43 of 54) did not result in a transaction or even in a rumor in the press. The remaining observation is a rumor on the boards that followed, rather than preceded, some press rumors.

This second analysis produces a different picture from the first. Boards quickly disseminate information and provide investors with a forum to digest it. They also share not-quite-yet-public information. But in our sample, the boards do not seem privy to truly material inside information. This observation is consistent with Glenn's observations and the empirical results on return predictability, which we discuss later.