Prior literature observes that information uncertainty exacerbates investor underreaction behavior. In this paper, I investigate whether, as professional investment intermediaries, sell‐side analysts suffer more behavioral biases in cases of greater information uncertainty. I show that greater information uncertainty predicts more positive (negative) forecast errors and subsequent forecast revisions following good (bad) news, which corroborates previous findings on the post‐analyst‐revision drift. The opposite effects of information uncertainty on forecast errors and subsequent forecast revisions following good versus bad news support the analyst underreaction hypothesis and are inconsistent with analyst forecast rationality or optimism suggested in prior literature.

We document strikingly opposite time‐series patterns of analyst forecast errors (FEs) and associated market reactions, illustrating that analyst forecasts have become a less useful benchmark of the market's earnings expectations in recent years. The mean FE has increased from negative one to two cents in the 1990s to positive one to two cents in the 2010s, whereas average earnings announcement returns have declined from 0.30% in the 1990s to −0.30% in the 2010s, turning negative in the past 17 years. Underlying the time‐series pattern of increasing FEs is a secular trend where firms move away from just meeting or beating, to which the market reaction has become increasingly negative, toward a large beat, while the frequency of meeting or beating the consensus analyst forecast remains stable during the same period. We develop a parsimonious predictive model of earnings surprises based on peer and past analysts' FEs and find that our predicted FE closely mirrors reported FE, with the average value hovering around one to two cents in most years of the past two decades. The market reaction to “around zero” unexpected FE (FE minus predicted FE) is indistinguishable from zero over time, suggesting that our model serves as a good benchmark of the market's expectation. Our evidence has broad implications for appropriate earnings benchmarking, for the disappearing discontinuity of the earnings surprise distribution around zero, for earnings management to beat analysts' forecasts, for empirical designs when examining the earnings‐return relation, and for the disappearing earnings announcement premium.