Bayesian MCMC methods of portfolio selection analysis

To solve the problem that portfolio selection is deeply influenced by the vector and covariance matrix of return, this paper present a Bayesian Markov Chain Monte Carlo (MCMC) algorithm which is used to estimate the mean vector and matrix of return. Comparison of the portfolio selection weight and between this approach and the sample mean and sample covariance of past log-returns, are studied on ten shocks historical data of SSE50 from 2007 to 2009. We carry out prediction of expected return by the last portfolio weight. Our method resulted in an extra 6.8 percentage points per year in additional portfolio performance which is quite a significant empirical result. Moreover, this approach has lower commission cost.