Deep Q-Networks for Reinforcement Learning in Trading Systems

Summary

The article explains Deep Q-Networks (DQN) as a reinforcement learning method that uses neural networks to estimate action values across market states. It outlines the agent, environment, state, action, and reward concepts, then describes a trading setup with buy, sell, or hold actions. DQN’s online and target networks estimate action values, while experience replay and discounted future rewards support learning. The article also contrasts this approach with table-based Q-learning, which can be less suited to large state spaces.

The implementation is presented as a signal component for an MQL5 Expert Advisor, where DQN supports the loss function of a separate MLP rather than acting as an independent signal generator. The reported test uses EURGBP daily data for 2023 and is described as demonstrating tradability, not future repeatability. The supplied text omits much of the implementation and the detailed tester results, so it does not allow independent assessment of performance, robustness, or generalization.

Key ideas

• DQN uses neural networks to estimate action values across environment states.
• The described trading agent chooses among buying, selling, and holding, with rewards used to update learning.
• An online network and a target network are used to estimate and train action values.
• In this implementation, DQN supports an MLP’s training loss instead of serving as a standalone signal generator.
• The EURGBP test is presented as illustrative and does not establish future performance.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.