Forecasting Stock Returns with Model Uncertainty and Parameter Instability
45 Pages Posted: 21 Sep 2017 Last revised: 11 Apr 2019
Date Written: February 12, 2019
We compare sophisticated models of forecasting stock returns that accommodate model uncertainty, including the new weighted-average least squares (WALS) technique. When estimated traditionally, our results confirm that the performance-based combination of individual predictors is superior. However, sophisticated models (especially WALS) improve dramatically once we combine them with the historical average and take parameter instability into account. Combining WALS with the historical average, for example, results in a statistically significant monthly out-of-sample R2 of 1.15% and annual utility gains of 2.55%. We obtain similar gains for predicting factor portfolios and macro economic conditions.
Keywords: Market Efficiency, Asset Pricing, Equity premia predictability, Forecast Combination, Model uncertainty, Parameter instability, WALS
JEL Classification: G17, G12, G02, C58
Suggested Citation: Suggested Citation