| Abstract [eng] |
In this paper, we examine reinforment learning methods and their sutability for use in stock trading automation by maximizing long term investment return. This paper consists 5 main chapters. In chapter 1 we take an overview of related works in the field. In chapter 2 stock trading problem we are trying to solve is defined. In chapter 3, the key concepts of reinforment learning is analyzed. In chapter 4 we analyze the problems when applying reinforment learning in complex systems. We analyze the extremum variants of reinforcement learning and examine modern deep reinforcement learning algorithms: Deep Deterministic Policy Gradient (DDPG), Twin Delayed DDPG (TD3) and Soft Actor Critic (SAC). In chapter 5, we propose software program for analysing automated stock trading using deep reinformecent learning techniques. We train a deep reinforcement learning models to optimize a stock portfolio containing 30 Dow Jones stocks. For the performance metrics anualized return measured by the Sharpe ratio is used. Results shows the TD3 algorithm suits the task better than DDPG and SAC. |
