Actor-Critic Methods in Stock Trading : A Comparative Study

Reinforcement learning has been proven to be extremely successful in a variety of fields, solving problems that were previously unsolvable. In our article, we apply this method of learning in the financial market, which is notorious for its complexity and stochasticity. Our objective is to find a strategy for the most effective negotiation of actions and consequently returns maximization. Therefore, we investigate the effectiveness of five reinforcement learning algorithms: Advantage Actor Critic (A2C), Deep Deterministic Policy Gradient (DDPG), Proximal Policy Optimization (PPO), Twin Delayed DDPG (TD3) and Soft Actor-Critic (SAC) when used as stock trading controllers. The results show that the 5 algorithms exceed the Dow Jones Industrial Average index with Proximal Policy Optimization (PPO) the best in terms of cumulative returns.