Validating Return-Generating Models

Performance measurement and event studies frequently assume a specific stochastic process for stock returns. The purpose of this paper is to validate the predictive accuracy of various stochastic processes on data different from those used in estimating the models. The main conclusion is that multi-factor models estimated with factor analytic techniques provide more accurate forecasts than the usual market model with either an equal- or value-weighted index, and Fama–French three-factor model. A model incorporating a set of pre-specified macro variables that some prior studies have used has no predictive value.