Reinforcement Learning Agents for Algorithmic Trading
In this project, I designed, implemented, and backtested a Deep Deterministic Policy Gradient (DDPG) Market Making Agent and a Deep Q Network (DQN) Market Taking Agent to explore innovative approaches in algorithmic trading. The Market Making Agent focused on earning profits from bid-ask spreads while managing inventory levels and mitigating risks, whereas the Market Taking Agent aimed to generate alpha by executing decisive buy and sell actions in the market. Strategy Studio was employed for backtesting these agents’ performance in simulated market environments.
Key contributions include:
- Developing a comprehensive state representation capturing essential market features such as bid-ask spread, order book imbalance, and volatility.
- Designing reward functions to reflect trading objectives and guide the learning process.
- Implementing neural network architectures for DQN and DDPG agents, including experience replay mechanisms to enhance learning efficiency.
- Utilizing Strategy Studio for backtesting and validating agent performance.
This project showcased a multidisciplinary approach, integrating deep learning, reinforcement learning, and quantitative finance methodologies to create and optimize advanced trading algorithms.