Enhancing Automated Trading with Sentiment Analysis: Leveraging Large Language Models for Stock Market Predictions

This study explores the use of Large Language Models (LLMs) for automating investment strategies through sentiment analysis of financial news, social media, and market data. By fine-tuning models like GPT-3 on financial datasets, sentiment indicators are extracted and integrated with traditional machine learning algorithms to predict stock price movements. A comparative analysis of various models, including LLM-based, traditional machine learning models, and hybrid approaches, was conducted. The results reveal that the hybrid model, combining LLM-generated sentiment with machine learning algorithms, outperforms other models in terms of both prediction accuracy and financial performance. The hybrid approach achieved an accuracy of 77.4%, cumulative returns of 17.2%, and a Sharpe ratio of 1.20, demonstrating its potential for real-world trading applications. These findings highlight the importance of sentiment data in enhancing market predictions and provide a promising framework for automating investment strategies. However, challenges such as ambiguity in sentiment classification and the need for model adaptation to changing market conditions remain. Future research should focus on improving sentiment analysis accuracy and incorporating reinforcement learning for real-time trading.