Mean-variance portfolios using Bayesian vector-autoregressive forcasts

Protfolio optimization is very sensitive to the forecats of returns and (co-)variances of the underlying assets. This paper applies a Bayesian vector-autoregression of the asset universe to predict the returns. Further, the co-variance matrix is forecasted by an Augmented GARCH estimation of the most volatile principle components of the return series. As an empirical illustration, the daily stock returns of the German stocks index DAX have been used to calculate some well-known mean-variance portfolios. Back-testing is used to evaluate the performance. The approach seems to be promising.