Model predictive control design for constrained Markov jump bilinear stochastic systems with an application in finance

In this study, we propose a solution to the model predictive control problem for a class of constrained discrete-time bilinear stochastic systems consisting of two coupled subsystems with Markov jumps. The first one includes a bilinear term in the state variables of the second subsystem and the input, whereas the second subsystem is described by a Markov switching vector autoregressive model. Furthermore, hard constraints imposed on the input manipulated variables. The results obtained are applied to the dynamic investment portfolio selection problem for a financial market with serially dependent returns and switching modes, subject to hard constraints on trading amounts. Our approach is tested on a real dataset from the New York Stock Exchange and the Russian Stock Exchange MOEX.