[3Xin2] Poster session 1

[3Xin2-80] Optimal execution strategy using Deep Q-Network with heuristics policy

〇Tatsuyoshi Ogawa¹, Kei Nakagawa², Kokolo Ikeda¹ (1.Japan Advanced Institute of Science and Technology, 2.Nomura Asset Management Co,Ltd.)

Keywords:Reinforcement Learning, Deep Learning

The optimal execution problem involves finding the optimal execution strategy that minimizes the cost of trading a specific volume of stocks within a certain period.
To address this, deep reinforcement learning methods including the Deep Q Network~(DQN) which approximates the action-value function Q through deep learning have been proposed for finding optimal execution strategies.
However, deep reinforcement learning faces challenges, such as instability in learning and the need for a huge amount of data.
Therefore, we propose incorporating strategies derived from insights of the financial field into conventional DQN methods during the learning process.
This approach is expected to be able to learn high-performing policies more stably.
Numerical experiments are conducted in environments with various noise tendencies to verify the effectiveness of the proposed method.
The results show that the proposed method can consistently reduce costs across all environments compared to baseline methods.