RAG-based system generates investor briefs from SEC filings and macro data using GPT-4o

A study examines the use of Large Language Models for fundamental analysis by processing company reports, macroeconomic data such as GDP and inflation, and U.S. Securities and Exchange Commission (SEC) documents from EDGAR. The system utilizes a Retrieval-Augmented Generation (RAG) approach to send preprocessed data via API to the GPT-4o model.

The researchers scanned data relevant to nine companies over four weeks, incorporating an exemplar investor knowledge document based on Kitchin cycles. This process produced automatic briefs for each company, which were then evaluated by nine individual investors regarding their usefulness.

The study aims to demonstrate the potential of LLMs in automating and enhancing the analysis of complex financial data sources.