Federated learning for algorithmic trading

• Main Author: Aggarwal, Anusha
• Other Authors: Anupam Chattopadhyay
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science
• Online Access: https://hdl.handle.net/10356/175265
• Institution: Nanyang Technological University

The advent of algorithmic trading has transformed the financial sector by introducing advanced models capable of executing high-speed transactions and strategies adept at interpreting market dynamics. Among these innovations, the integration of Deep Reinforcement Learning (DRL) has been a major breakthrough. However, its reliance on extensive and diverse training datasets presents challenges in adaptability and generalization, especially when data availability is constrained by privacy concerns and market volatility. This study introduces an advanced federated learning framework tailored to algorithmic trading, which utilises the benefits of decentralised data training while maintaining confidentiality and leveraging collective intelligence. By simulating diverse market conditions through various trading and time costs across agents, this project examines the efficacy of federated learning against traditional non-federated training methodologies. The comparison of these paradigms on a consistent dataset underscores the potential for federated learning to enhance the robustness and performance of trading algorithms. The outcomes indicate that federated learning holds promise for developing resilient trading strategies, paving the way for more adaptable and scalable algorithmic trading solutions in the financial industry's increasingly complex landscape.