On the Economic Significance of Stock Return Predictability

Abstract

Abstract We study the effects of time-varying volatility and investment horizon on the economic significance of stock market return predictability from the perspective of Bayesian investors. Using a vector autoregression framework with stochastic volatility (SV) in market returns and predictor variables, we assess a broad set of twenty-six predictors with both in-sample and out-of-sample designs. Volatility and horizon are critically important for assessing return predictors, as these factors affect how an investor learns about predictability and how she chooses to invest based on return forecasts. We find that statistically strong predictors can be economically unimportant if they tend to take extreme values in high volatility periods, have low persistence, or follow distributions with fat tails. Several popular predictors exhibit these properties such that their impressive statistical results do not translate into large economic gains. We also demonstrate that incorporating SV leads to substantial utility gains in real-time forecasting.