Dynamic Portfolio Optimization and Economics Uncertainties

The development and use of dynamic optimization model is extremely important in financial markets. The classical mean-variance portfolio model assumes the expected returns are known with perfect precision. In practice, however, it is extremely difficult to estimate precisely. While portfolios that ignore estimation error have very poor properties: the portfolio weights have extreme values and fluctuate dramatically over time. The Bayesian approach that is traditionally used to deal with estimation error assumes investors have only a single prior or is neutral to the risk. Further, the Bayesian approach has computational difficulty to incorporate future uncertainty into the model. In this paper, I introduce Genetic algorithms technique in solving a dynamic portfolio optimization system, which incorporate economic uncertainties into a state dependent stochastic portfolio choice model. The advantage of GA is that it solves the model by forward-looking and backward-induction, which incorporates both historical information and future uncertainty when estimating the asset returns. It significantly improves the accuracy of mean return estimation and thus yields a superior model performance compared to the traditional methodologies. The empirical results showed that the portfolio weights using the GA model are less unbalanced and vary much less over time compared to the mean-variance portfolio weights. GA achieves a much higher Sharpe ratio and the out of sample returns generated by the GA portfolio model have a substantially higher mean and lower volatility compared to the classical mean-variance portfolio strategy and Bayesian approach.