Measuring Herding and Exaggeration by Equity Analysts and Other Opinion Sellers

Firms and individuals who sell opinions may bias their reports for either behavioral or strategic reasons. This paper proposes a methodology for measuring these biases, particularly whether opinion producers under or over emphasize their private information, i.e. whether they herd or exaggerate their differences with the consensus. Applying the methodology to I/B/E/S analysts reveals that they do not herd as is often assumed, but rather they exaggerate their differences with the consensus by an average factor of about 2.4. Analysts also overweight their prior-period private information and thus under-update based on last period's forecast error; this under-updating helps explain the apparently conflicting over and under-reaction results of DeBondt and Thaler (1990) and Abarbanell and Bernhard (1992). A useful by-product of the methodology is a measure of the incremental information content of an analyst's forecasts. Using this measure reveals that analysts differ greatly in performance: the information content of the future forecasts of the top 10 percent of analysts is roughly six times that of the bottom 40 percent.