Sell-side earnings estimates exhibit herding behaviour that is well documented in the academic literature but underappreciated in practice. Analysts revise toward the consensus rather than toward their own private signal when the social cost of being wrong alone exceeds the social cost of being wrong with the herd. The result is a consensus that is systematically biased in identifiable ways.
Where herding is strongest
Our analysis of 12,000 earnings estimate revisions across five sectors shows that herding is most pronounced in the two weeks before earnings — when the social cost of a bold revision is highest and the time to be vindicated is shortest. It is also stronger in covered companies with fewer than eight analysts, where individual outlier estimates are more visible.
A framework for fading consensus
Three conditions together are associated with consensus estimates that are biased by herding: high analyst turnover in the covered company in the preceding 12 months, a recent management guidance event that all analysts attended, and a narrow estimate range relative to historical volatility. When all three are present, the actual result deviates from consensus by more than one standard deviation in 61% of cases.