The Restrictions on Predictability Implied by Rational Asset Pricing Models

This article shows how rational asset pricing models restrict the regression-based criteria commonly used to measure return predictability. Specifically it invokes no-arbitrage arguments to show that the intercept, slope coefficients, and $$R^2$$ in predictive regressions must take specific values. These restrictions provide a way to directly assess whether the predictability uncovered using regression analysis is consistent with rational pricing. Empirical tests reveal that the returns on the CRSP size deciles are too predictable to be compatible with a number of well-known pricing models. However, the overall pattern of predictability across these portfolios is reasonably consistent with what we would expect under circumstances where predictability is rational.