Modeling the Cross Section of Stock Returns: A Model Pooling Approach

Abstract

Model selection, i.e., the choice of an asset pricing model to the exclusion of competing models, is an inherently misguided strategy when the true model is unavailable to the researcher. This paper illustrates the advantages of a model pooling approach in characterizing the cross section of stock returns. The optimal pool combines models using the log predictive score criterion, a measure of the out-of-sample performance of each model, and consistently outperforms the best individual model. The benefits to model pooling are most pronounced during periods of economic stress and it is a valuable tool for asset allocation decisions.

Keywords: Asset pricing, Model pooling, Model combination, Forecasting, Predictive distributions, Log predictive score

JEL Classification: G12, C52, C53