Reading the Tea Leaves: Why Serial Correlation Patterns in Analysts' Forecast Errors are not Evidence of Inefficient Information Processing

Abstract:     
This paper argues that an absence of serial correlation in forecast errors is not the appropriate benchmark for rational analyst behavior. We put forward a model that confronts analysts with two layers of uncertainty. An initial layer of uncertainty about firm-specific parameters leads analysts to underreact to signals from some firms and overreact to signals from others. A subsequent layer of uncertainty about the distributions from which these firm-specific parameters are drawn causes the null hypothesis of serially uncorrelated forecast errors to be frequently rejected despite being true. We then test our learning model’s predictions using IBES data, finding support for the view that analysts learn about individual firms in the face of time-varying model uncertainty.

Number of Pages in PDF File: 49

Keywords: Parameter uncertainty, learning, financial analysts

JEL Classification: G14, G24

working papers series