DTSL-PPO

Introduction

• This is the code repository for the solution 'DTSL-PPO: A Well-Tailored Deep Reinforcement Learning Model to Highly Profit in Futures Trading Market' to 'Trading Strategies', problem C of MCM 2022.
• This solution was a Finalist (Top 2%) Awardee.

Overview

• We modeled a series of trading conditions and operations as Markov Decision Process (MDP) and innovatively trained the trading agent with PPO algorithm based on deep reinforcement learning (DRL);
• We designed Dynamic T-period Sliding Window PPO (DTSL-PPO) algorithm with trading agents focusing respectively on long and short terms coordinating together, so as to initiate trading quickly with a relatively insufficient amount of data and perceive the market more precisely;
• We gave analyses to the performance of our model. The performance of agents trained by our algorithm was satisfying.

Framework