Deep Reinforcement Learning for Optimizing Finance Portfolio Management

Deep reinforcement learning (DRL) is an emerging artificial intelligence (AI) research field which combines deep learning (DL) for policy optimization and reinforcement learning (RL) for goal-oriented self-learning without human intervention. We address major research issues of policy optimization for finance portfolio management. First, we explore one of the deep recurrent neural network (RNN) models, GRUs, to decide the influences of earlier states and actions on policy optimization in non-Markov decision processes. Then, we craft for a viable risk-adjusted reward function to evaluate the expected total rewards for policy. Third, we empower the integration of RL and DL to leverage their respective capabilities to discover an optimal policy. Fourth, we investigate each type of RL approaches for integrating with the DL method while solving the policy optimization problem.