Portfolio Allocation and Reinforcement Learning

In this chapter, we briefly review the methodology of reinforcement learning and describe its application to the financial problem of portfolio allocation. In this context, we define the environment as a set of states, captured by such financial variables as stock returns or technical indicators, and of actions, mainly the determination of wealth shares to invest in each asset. Optimal value functions are obtained through the Bellman optimality equation, a well-established principle in both reinforcement learning and portfolio optimization. Deep reinforcement learning algorithms have the advantage of providing approximate solutions since most portfolio problems lack analytical solutions. We describe several algorithms and apply them to classical portfolio allocation problems, where risk minimization and return maximization are combined with or without accounting for trading costs.