Reinforcement Learning for Algorithmic Trading in Financial Markets

In multi-period algorithmic trading, determining algorithms that are ideal for risk-averse strategies is a challenging task. This study explored the application of model-free reinforcement learning (RL) in algorithmic trading and analyzed the relationship between risk-averse strategies and implementation of RL algorithms including Q-learning, Greedy-GQ, and SARSA. The data for this quantitative research included one year of the E-mini-NASDAQ-100 futures (2023-2024). Over 7,500 simulation results substantiated a proof of concept that Q-learning can successfully generate risk-adjusted trading signals in the highly liquid technology-focused futures market. With an optimized configuration of hyperparameters including look-back period, basis and reward functions, Q-learning delivered nearly twice the returns of the competing RL algorithms. Beyond absolute returns, Q-learning exhibited lower volatility across key risk metrics and outperformed the NASDAQ-100 benchmark by approximately 75 percentage points. These findings suggest reinforcement learning as a promising artificial intelligence and machine learning framework for alpha generating strategies in systematic trading.

Author

• Gityforoze, Soheil

Language

• en

Keyword

• Reinforcement Learning
• Financial Markets
• Hyperparameter Optimization
• Q-Learning
• Machine Learning
• Algorithmic Trading

Date created

• 2025

Type of Work

• Dissertation

License

• All rights reserved

Rights statement

• In Copyright

GW Unit

• Systems Engineering

Degree

• D.Engr.

Advisor

• Focacci, Aaron A.F.
• Raghavan, Vijay V.R.

Committee Member(s)

• Alsarhan, Hamza H.A.

Persistent URL

•  https://scholarspace.library.gwu.edu/etd/2j62s5932