BERT for Stock Trading: Can Large Language Models Predict Long-Term Stock Price Movements?

To answer whether large language models can predict stock price movements over the long-term, such as a one-year time period, the authors fine-tune transformer-based large language models, including BERT, on different sources of company-related text data for a classification task to predict the one-year stock price performance. They use three different types of text data: news articles, financial blogs, and annual reports. This allows them to analyze to what extent the performance of large language models is dependent on the type of the underlying document. Their transformer-based stock performance classifier shows substantial improvement in predictive accuracy compared to traditional language models. The highest performance was achieved with news articles as text sources, suggesting that the effectiveness of large language models for the prediction task is likely reliant on the underlying quality of the text data. Performance simulations indicate that these improvements in classification accuracy also translate into above-average stock market returns.