Stock return predictability: riding the risk premium



Publishing details

Social Sciences Research Network

Authors / Editors

Gomez-Cram R


Publication Year



Past returns contain rich information about future returns. I propose an approach to estimate expected returns based on the full history of past returns, which is able to outperform the prevailing mean benchmark on a consistent basis, over long sample periods, and with monthly out-of-sample R2 statistics of at least 2% and annualized utility gains greater than 300 basis points. My approach allows for correlated shocks between unexpected and expected returns and ties expected return variations to business-cycle fluctuations. These properties generate different persistence, volatility, and serial correlation of expected returns across economic states and determine how the information in lagged returns is used to predict future returns. My approach has important implications for standard predictive regressions.

Series Number



Social Sciences Research Network