Stock-Return Predictability and Asset Pricing Models

The regression of stock returns on predictive variables, such as dividend yield, has proven useful in optimal portfolio selection when investment opportunities are timevarying. Conditional versions of factor models impose a restriction on that regression, thereby implying a particular portfolio choice. The study examines several pricing models from a perspective of conditional mean-variance optimizing investors. We show that the tangency portfolios based on predictive regressions that conform to the Fama and French (1993) restriction and factor models that account for the momentum anomaly of Jegadeesh and Titman (1993) and other exposures, differ from their unrestricted counterparts to economically significant amounts. Examining the restriction on predictability implied by the Daniel and Titman (1997) model, we find that optimal portfolios based on restricted and unrestricted predictive regressions differ to significant degrees as well. The results carry implications for inference in long horizon event studies, evaluating equity mutual funds and hedge funds, and other applications.