Optimal Execution of a Basket of Stocks under Regime-Switching Model Using Reinforcement Learning

In this article, the authors are concerned with the problem of successfully trading a portfolio of stocks when the overall marketplace is subject to change between various trends during a relatively short period. Such executions usually involve large positions of highly correlated assets. In this article, they address this problem as an optimal control of a multidimensional Markov switching process. They propose a solution based on a reinforcement learning methodology. They prove the convergence of their approximating scheme. They present an implementation of their algorithm using intraday data.