Factor Covariances Predict Factor Returns

Abstract

We examine low-turnover zero-investment "factor" portfolios constructed from various stock characteristics previously shown to predict returns. The nine different factor portfolios all exhibit negative market betas. Our central result is that a more negative beta across factors predicts higher factor returns over the next two years. Similarly, the average relative volatility of the factor returns, as well as the cross-sectional variance of the betas and volatilities, predicts future factor returns. While the results are difficult to reconcile with standard risk-based explanations, they are consistent with the existence of a time-varying mass of naive investors, whose trading affects the returns to characteristics-based factor portfolios. Indeed, the average beta across factors is highly negatively correlated across time with the Baker and Wurgler (2006) investor sentiment measure.

Keywords: Factor Returns, Factor Covariances, Investor Sentiment, Time-Series Predictability

JEL Classification: G12, G02