We show analytically that mean analyst forecasts inefficiently aggregate information by assigning too much weight to analysts’ common information relative to their private information when used as a summary forecast measure of forthcoming earnings. A more precise summary forecast of earnings than the current mean forecast is the current mean forecast plus a positive multiple of the change in the mean forecast.

In this study we examine changes in the precision and the commonality of information contained in individual analysts' earnings forecasts, focusing on changes around earnings announcements. Using the empirical proxies suggested by the Barron et al. (1998) model that are based on the across-analyst correlation in forecast errors, we conclude that the commonality of information among active analysts decreases around earnings announcements. We also conclude that the idiosyncratic information contained in these individual analysts' forecasts increases immediately after earnings announcements, and that this increase is more significant as more analysts revise their forecasts. These results are consistent with theories positing that an important role of accounting disclosures is to trigger the generation of idiosyncratic information by elite information processors such as financial analysts (Kim and Verrecchia 1994, 1997).

[This paper presents a model that relates properties of the analysts' information environment to the properties of their forecasts. First, we express forecast dispersion and error in the mean forecast in terms of analyst uncertainty and consensus (that is, the degree to which analysts share a common belief). Second, we reverse the relations to show how uncertainty and consensus can be measured by combining forecast dispersion, error in the mean forecast, and the number of forecasts. Third, we show that the quality of common and private information available to analysts can be measured using these same observable variables. The relations we present are intuitive and easily applied in empirical studies.]

This paper presents a model that relates properties of the analysts' information environment of the properties of their forecasts. First, we express forecast dispersion and error in the mean forecast in terms of analyst uncertainty and consensus (that is, the degree to which analysts share a common belief). Second, were reserve the relations to show how uncertainability and consensus cab be measured by combining forecast dispersion, error in the mean forecast, and the number of forecasts. Third, we show that the quality of common and private information available to analysts can be measured using these same observable variables. The relations we present are intuitive and easily applied in empirical studies.