|
Abstract:
|
Using Bayesian dynamic linear models, we provide a method of predicting risk premia that incorporates time-varying parameters, stochastic volatility and variance discounting. These models are capable of jointly predicting the risk premia for multiple assets which permits the strategic allocation of these assets in a portfolio. This methodology is demonstrated for a portfolio containing a stock index, bond index, and a risk-free asset. We average across many models with different combinations of predictors and discount factors. Our averaged models show statistical and economic improvements in out-of-sample predictability compared to models incorporating subsets of the aforementioned features as well as the baseline historic mean model.
|
