Model predictive control for portfolio selection

Abstract

In this paper, we explain the application of model predictive control (MPC) to problems of dynamic portfolio optimization. At first we prove that MPC is a suboptimal control strategy for stochastic systems which uses the new information advantageously and thus, is better than pure optimal open-loop control. For a linear Gaussian factor model, we derive the wealth dynamics and the conditional mean and variance. We state the portfolio optimization, where an investor maximizes the mean-variance objective while keeping the portfolio value-at-risk under a given limit. The portfolio optimization is applied in a case study to US asset market data.