This paper introduces GPT-InvestAR, a novel framework that enhances stock investment strategies by automating the analysis of Annual Reports using Large Language Models (LLMs). Annual Reports are critical for assessing a company's financial health but are often lengthy and complex, making manual analysis time-consuming and requiring expert knowledge. The authors address this challenge by leveraging LLMs to extract valuable insights from these documents, compiling the outputs into a quantitative dataset that is augmented with historical stock price data. This approach transforms unstructured textual information into structured features suitable for machine learning applications.

A Machine Learning model is then trained using these LLM-generated features to predict stock performance, with the methodology validated through walkforward testing. The results demonstrate promising outperformance compared to S&P500 returns, indicating that the framework can effectively enhance investment decision-making. By releasing the code as open source, the paper provides a scalable and accessible tool for future research in AI-driven financial analysis, potentially democratizing access to sophisticated investment strategies and reducing reliance on manual expertise.