Updating Expected Returns Based on Consensus Forecasts

Abstract

Investor behavior can explain to some extent the stock market anomalies from a psychological viewpoint. Recent literature suggests a lot of models without testing predictability implied by the models and without a discussion of implications and limitations that are implied by the design. Mostly, these models are descriptive. In these designs, the question about relevant normative models is left aside. In this paper we propose a normative model that allows empirical testing of whether the way investors should behave given the information is useful in making judgments in financial markets. Contrary to most papers, we apply individual priors to form a judgment about the future price change of each asset at each point in time. These priors are considered as the expert opinion and are given by the one-year conensus forecast of earnings yield as provided by analysts. This design allows tests of the predictions for a normative setting using actual market data. Comparing Bayes' rule to a decisions by a price trader, we find that economic loss is lower for the price trader than for the Bayesian trader under several specifications. However, using expert information in the Bayes' rule leads to better predictions for stocks that do not have high-risk characteristics.

Keywords: Bayesian Simulation, Bayes' Rule, normative models, expected returns

JEL Classification: C11, C15, G12