"Model-Uncertainty for Long-Term Investors"


Speaker


Abstract

We develop a method to incorporate model uncertainty with respect to restricted VAR(1) models using Bayesian Model Averaging (BMA) and apply our method to analyze the long-run predictability of asset returns. We find that only the dividend yield and credit spread are important predictors of stock returns in the short run, but that almost all considered predictors are important for long-run predictability. Next, the analysis shows that model uncertainty is substantial, since the posterior probability is widely spread among many models. Furthermore, despite clear evidence of mean-reversion in stock returns, we show that stocks are in general at least as risky in the long-run as in the short-run and that stocks are even riskier in the long-run in case of an economic crisis such as the recent subprime mortgage crisis. Finally, the strategic asset allocations for longterm investors using BMA are substantially different from investors that use the highest posterior probability model. Our analysis relates this finding to a lower mean, higher variance, more negative skewness and a higher kurtosis of the predictive distributions of excess stock returns when incorporating model uncertainty. Differences are especially large when the economy deviates substantially from its steady state value.
 
Contact information:
Kees Bouwman
Email