Improving equity premium forecasts by incorporating structural break uncertainty

Abstract

This paper compares five alternative methods for directly dealing with structural break uncertainty in forecasting the U.S. equity premium using 30 widely used bivariate and multivariate predictive regressions. We find that two recently developed methods -Robust Optimal Weights on Observations and Forecast Combination across Estimation Windows -outperform the conventional rolling window and post-break estimation methods. This result indicates that very early historical information is beneficial for U.S. equity premium forecasting but should be discounted to incorporate structural break uncertainty.