GPT-4o RAG System Automates Investor Briefs from SEC Filings
New research feeds GPT-4o SEC filings and macro data for automated analysis.
A new paper from researchers Bartosz Ziółko and Kacper Dobrzeniewski explores using large language models (LLMs) to augment fundamental company analysis. They developed a Retrieval-Augmented Generation (RAG) system powered by OpenAI's GPT-4o, which ingests data from SEC EDGAR filings and macroeconomic indicators (GDP, inflation) to produce automated investor briefs. The system also incorporates Kitchin cycles—short-term economic cycles of 3–5 years—to provide macro context. The study involved preprocessing these data sources and sending them via API to GPT-4o in a RAG-like regime.
The system was tested on 9 companies over a 4-week period, generating automatic briefs that were evaluated by 9 individual investors. The goal was to assess whether LLM-generated summaries could replace or augment manual analysis. While the paper is pending publication, it signals a growing trend: using AI to process dense financial documents at scale. For professionals, this could mean faster screening of investment opportunities, though the reliability and depth of AI-generated briefs remain under scrutiny. The authors are from the academic community and have made the paper available on arXiv.
- System uses GPT-4o via API with a RAG architecture to process SEC EDGAR filings and macroeconomic data.
- Tested on 9 companies over 4 weeks, with 9 individual investors evaluating the automated briefs.
- Incorporates Kitchin cycles (3–5 year economic cycles) for additional macro context in analysis.
Why It Matters
Automating SEC filing analysis with LLMs could save investors hours of manual work each week.