Modeling limit order trading with a continuous action policy for deep reinforcement learning.

Area of Science:

• Quantitative Finance
• Computational Finance
• Machine Learning

Background:

• Market orders in algorithmic trading risk slippage, leading to unfavorable execution costs.
• Limit orders offer price protection but pose execution risks and complexity, limiting their use in machine learning (ML) trading systems.
• Current ML trading literature predominantly utilizes market orders, neglecting the benefits of limit orders.

Purpose of the Study:

• To develop a Deep Reinforcement Learning (DRL) framework enabling trading agents to effectively utilize limit orders.
• To address the challenge of limit order execution risk within ML-based trading strategies.
• To integrate a hybrid approach allowing both limit and market orders based on risk assessment.

Main Methods:

• A novel DRL approach is proposed, modeling limit prices using a continuous probability distribution.
• The framework incorporates a mechanism to switch to market orders when the risk of non-execution outweighs slippage costs.
• Extensive experiments were performed on multiple currency pairs with hourly price data.

Main Results:

• The proposed DRL method demonstrates effectiveness in modeling limit order behavior within trading agents.
• The hybrid order strategy proved advantageous in managing execution risk and slippage.
• Experimental validation confirmed the method's viability across various currency pairs.

Conclusions:

• The developed DRL approach successfully incorporates limit orders into automated trading systems.
• This research opens new avenues for applying DRL to financial markets utilizing limit order strategies.
• The findings suggest a significant advancement in ML-based algorithmic trading by enabling sophisticated order type modeling.