Understanding Bias and Dispersion in Forecasts: The Role of Motivated Reasoning

Abstract

This paper reports the results of an experiment showing that individuals who have stakes in the earnings performance they are forecasting tend to generate forecasts that are more dispersed and more biased when the news they are evaluating indicates earnings performance inconsistent with their preferences. These effects work partly through participants' beliefs about potential bias in the news they evaluate. Participants who prefer lower earnings are more skeptical of optimistic bias in news and provide lower forecasts than participants who prefer higher earnings. After controlling for the effect of participant beliefs about biased news, the forecasts of participants with a preference for lower (higher) earnings performance are still relatively pessimistic (optimistic). Forecasters also take the preference-consistent news more at face value, whereas they rely relatively more on their idiosyncratic beliefs when news is preference inconsistent. Because beliefs about optimistic bias differ across individuals, preference-inconsistent news induces more dispersed forecasts by magnifying that source of disagreement. Because their beliefs are influenced by the direction of their earnings preferences, preference-inconsistent news also induces more biased forecasts by magnifying that source of bias. Stakes, therefore, lead to predictable bias and dispersion. This bias-dispersion link provides an ex ante predictor of bias in forecasts because forecast dispersion is immediately observable, whereas bias is only measurable ex post. These results shed light on the causes of dispersion in forecasts and suggest directions for analytical models of heterogeneous beliefs.