More factors matter and factors matter more than you might think: The role of time variation in factor premia

Abstract

The literature has asserted that as few as four or five factor principal components (PCs) are sufficient to largely explain the cross-section of stock returns. Allowing for time variation in factor premia, we show that portfolios formed from factor PCs yield economically large out-of-sample Sharpe ratios that increase on average until about forty PCs are employed. That is, non-latent time-varying factors have strong predictive power for the cross-section of stock returns, and to a substantial extent are not redundant of each other. Time variation in the number of economically relevant factors is related to changes in economic conditions and the diversity of firm characteristics. These results imply that barriers to cross-factor arbitrage are large and indicate roles for economic complexity and investor learning in asset pricing.

Keywords: Asset Pricing, Factors, Complexity, Time series variation, Diversity

JEL Classification: G10, G11, G12