期货合约
经济
强化学习
计算机科学
控制(管理)
微观经济学
交易策略
业务
期货市场
期限(时间)
订单(交换)
计量经济学
算法交易
出处
期刊:
[Institute of Electrical and Electronics Engineers]
日期:2026-01-29
卷期号:: 1-6
标识
DOI:10.1109/icses66558.2026.11479126
摘要
The trading of the futures markets based on high-frequency trading (HFT) demands high-speed trading with very dynamic and stochastic conditions, which cannot allow using the traditional rule-based approach as the one that is profitable in the long term. The article is a high frequency trading framework optimization article of the Chinese CSI 300 futures market by a Deep Q-Network (DQN)-based Reinforcement Learning (RL) algorithm. The proposed solution is a Markov Decision Process to trade the environment, and the market conditions are constructed using high-frequency order book properties, price momentum indicators, and volatility indicator. By means of continuous interaction with the market environment, the agent, which is DQN, can learn the best activities in trading so that it can be able to buy, trade, or hold. To make the process more convergent and stable, experience replay and target network are introduced in the learning. The market effects, transactions costs and slippage are explicitly considered so as to make the trading behavior realistic. Extensive simulations on historical data of CSI 300 futures index-based outperform the proposed RL-based strategy, with a greater profitability, Sharpe ratio, and drawdown than the normal HFT benchmarks. The results confirm that deep reinforcement learning can be used to acquire complex market microstructure patterns, and real-time trading strategy. The study provides practical information as far as the designing of smart trading systems in new futures markets are concerned.
科研通智能强力驱动
Strongly Powered by AbleSci AI