Researchers propose RAG-based system to automate investor briefs using company filings and macroeconomic data

A new arXiv preprint introduces a Retrieval-Augmented Generation (RAG)-based system designed to automate the creation of investor briefs by synthesizing company-specific reports, U.S. Securities and Exchange Commission (SEC) filings from EDGAR, and macroeconomic indicators such as GDP and inflation data.

The authors implemented a pipeline that preprocessed these data sources and transmitted them via API to the GPT-4o model in a RAG-like configuration. They also incorporated a document outlining exemplar investor knowledge grounded in Kitchin cycles to contextualize the analysis.

Over a four-week period, the system generated weekly briefs for nine companies. These briefs were then distributed to nine individual investors, who assessed their usefulness as part of the study’s evaluation framework.

The study situates its contribution at the intersection of computational linguistics and financial analysis, proposing that LLMs can reduce manual effort in fundamental analysis while maintaining or improving the quality of insights delivered to investors.