Equity premium predictability using Bayesian regression mixtures

This study investigates the predictability of equity returns in a mixture modelling framework that address several potential shortcomings of standard regression-based tools. We find that static single-equation predictive models are strongly rejected by the data in light of 2 and 3-component mixture regression specifications. We generalise standard regression mixtures to accommodate the possibility that predictive variables can be used to forecast time-variation in mixing weights. While within-sample variation in mixing weights appear forecastable, we find little benefit to forecasting such variation on an out-of-sample basis.