Reinforcement learning in Algorithmic Trading

Bold-Erdene Bayaraa

Bachelor's thesis

Abstract:

This thesis investigates the application of deep reinforcement learning (DRL) for the development of algorithmic trading strategies. Recognizing the limitations of traditional methods that often rely on fixed financial models, this research explores the potential of DRL to learn and adapt directly from market data. The study centers on a comparative analysis of five DRL algorithms (A2C, DDPG, PPO, TD3, SAC), benchmarked against the DJIA. Using historical market data within a FinRL-powered trading environment, these agents are trained and evaluated based on core financial metrics like Sharpe ratio, drawdown, and risk-adjusted returns. A significant finding is the adaptability of DRL strategies, with the DDPG agent demonstrating particular resilience during market downturns. This outperformance highlights the potential of DRL to surpass traditional index-tracking approaches in volatile conditions. While the PPO algorithm exhibited weaker performance, the overall results showcase the promise of DRL for algorithmic trading. However, the complexity of real-world markets warrants further investigation. Future research should address limitations such as transaction costs, market impact, and the integration of more diverse data sources. Additionally, advancements in explainable RL could enhance the transparency and trustworthiness of these systems within the financial domain. This thesis contributes to the growing field of research exploring the intersection of machine learning and finance. By providing insights into the strengths, weaknesses, and practical implementation considerations of DRL algorithms for trading, this work lays a foundation for the development of increasingly intelligent and robust algorithmic trading strategies.

Keywords

Reinforcement Learning, Deep Reinforcement Learning, Machine Learning, Algorithmic Trading

Language used: English

Date on which the thesis was submitted / produced: 18. 3. 2024

Thesis defence

• Date of defence: 20. 5. 2024
• Supervisor: Ing. Martin Pelikán, Ph.D.
• Reader: Martin Čejka, Ing.

Czech University of Life Sciences Prague

Faculty of Economics and Management

Bachelor programme:
Informatics