Financial trading in the digital age: the integration of large language model and reinforcement learning

Abstract

In recent years, quantitative trading has gained significant traction in the financial markets. The traditional strategies primarily rely on mathematical and statistical models, while a growing number of hedge funds have begun to explore machine learning-based algorithms, for developing sophisticated trading strategies. However, these purely quantitative strategies, which solely utilize data such as price and order-book information, may underperform when faced with market fluctuations triggered by newly released information, such as policy changes, news updates, and financial reports.

This project proposes a novel system that effectively integrates reinforcement learning methods with large language models for a comprehensive analysis of both quantitative data and real-time market sentiment. Reinforcement learning is adept at identifying patterns in quantitative markets, while large language models excel at processing and summarizing recent online information to gauge market sentiment. This innovative blend of digital and market insights provides a well-rounded strategy to navigate immediate market trends and potential fluctuations, aiming to offer a more holistic and adaptive approach to market analysis and decision-making.