Return Predictability under Equilibrium Constraints on the Equity Premium

This paper proposes a new approach for incorporating theoretical constraints on return forecasting models such as non-negativity of the conditional equity premium and sign restrictions on the coefficients linking state variables to the equity premium. Our approach makes use of Bayesian methods that update the estimated parameters at each point in time in a way that optimally exploits information in these constraints. Using a variety of predictor variables from the literature on predictability of stock returns, we find that theoretical constraints have an important effect on the coefficient estimates and can significantly reduce biases and estimation errors in these. In out-of-sample forecasting experiments we find that models that exploit the theoretical restrictions produce better forecasts than unconstrained models.