Predicting Returns Out of Sample: A Naïve Model Averaging Approach

Abstract

Prior literature finds that variables that can forecast market returns in sample do not beat historical averages in forecasting market returns out of sample. We propose a naïve model averaging (NMA) method, which produces mostly positive out-of-sample R2s for the variables that are significant in sample. The NMA method is also helpful relative to other more sophisticated methods. Surprisingly, more sophisticated weighting schemes that combine the predictive variable and the historical mean do not consistently perform better than the NMA method. Model misspecification, rather than declining return predictability, likely explains the predictive performance of the NMA method.

Keywords: out of sample tests, naïve model averaging, market returns, ridge regression

JEL Classification: G12, G11