Return Predictability, Economic Profits, and Model Mis-Specification: How Important are the Better Specified Models?

摘要

This paper addresses the question whether investors can profit from return predictability in the real world while focusing on the impact of the data-generating process (DGP). We estimate an array of predictive models ranging from the simplest VAR to nonparametric ones and evaluate their out-of-sample portfolio performance with various predictive variables. We find that despite the significant statistical improvement the better specified predictive models do not consistently outperform the VAR. Another striking finding is that investors appear to be better off predicting only the sign but not the magnitude of the market expected excess returns.