Trading Gold, GBPUSD and EURUSD with Deep Reinforcement Learning Agents

Trading in the financial markets is driven by both fundamental and technical analysis. These methods have however been shown to have limitations owing to the inefficiency of the markets attributed to human behavior and psychology. Algorithm led methods have been highlighted as a suitable alternative which for years has banked on simple heuristics and rules. Deep Reinforcement Learning presents the next frontier in these automated methods owing to its ability to develop agents that can master any given environment like financial markets. In this paper, these agents are tested by trading three financial instruments. While developing these agents, simple prediction models are first implemented using a variety of deep learning algorithms. Sequence to sequence models outperform all other algorithms in the prediction of the three instruments. As for the agents, their performance is not consistent across the three instruments but Gold is traded with better results that is, better profit and loss measures. In the end, the paper recommends the use of better trade management heuristics to determine quality of trades taken and the refinement of the action space to incorporate risk management strategies.