[2J4-GS-2] Machine learning: Deep reinforcement learning

[2J4-GS-2-03] Compound deep reinforcement learning to acquire trading strategies in a complex environment

〇Takuma Kondo¹, Tohgoroh Matsui¹ (1. Chubu University)

Keywords:Reinforcement learning, finance

This paper proposes a method to extend the latest deep reinforcement learning algorithm to a compound deep reinforcement algorithm and to acquire a financial trading strategy in more complex environments by adding the state variables.

Previous research used only two state variables while one of the good points of deep reinforcement learning is that it can use many state variables.

And it used a compound deep reinforcement learning algorithm based on the simplified DQN that is the earliest deep reinforcement learning algorithm.

In this paper, we extend the latest algorithms of deep reinforcement learning to a compound type and acquire a financial trading strategy in a complex environment represented by many state variables.

In addition, we applied the proposed method to acquire a trading strategy for Japanese government bonds and confirmed its effectiveness.