Harnessing Federated Learning for Stock Analysis and Portfolio Optimisation

Author

Nishanth Sagar Panthangi, Southern Illinois University CarbondaleFollow

Date of Award

12-1-2024

Degree Name

Master of Science

Department

Computer Science

First Advisor

Ahmed, Dr. Khaled

Abstract

Financial technology is rapidly transforming, with machine learning becoming crucial in stock analysis and portfolio optimization. Traditional stock analysis faces data privacy challenges when pooling data. Federated Learning addresses this by training models collaboratively without sharing raw data. This thesis explores Federated Learning in stock analysis, focusing on optimizing portfolio strategies using client-specific data.The proposed Federated Learning framework allows entities to develop independent models while maintaining data privacy. By combining diverse trading strategies, a global model is built, enhancing adaptability and predictive accuracy. The global model achieved a 25.08% CAGR and a Sharpe ratio of 4.46, outperforming individual models, which had Sharpe ratios from 1.75 to 2.67 and CAGR values from 12.5% to 18.3%. This demonstrates Federated Learning's potential in financial modeling, enabling shared insights without compromising data security. This study exhibits Federated Learning's potential for scalable, privacy-focused collaborative stock analysis. By bridging predictive capabilities with privacy, Federated Learning can transform financial decision-making, yielding higher risk-adjusted returns. The federated model's superior Sharpe ratio and reduced drawdown illustrate its effectiveness. The implementation involved local models being aggregated into a global model using federated averaging, resulting in improved performance metrics across varying market conditions.