FinRpt: A New Benchmark and Multi-Agent Framework for Automated Equity Research Report Generation

The introduction of FinRpt, a novel open-source benchmark, aims to enhance the automated creation of Equity Research Reports (ERRs) utilizing large language models (LLMs). This initiative, outlined in arXiv paper 2511.07322, is the first to define the ERR Generation task, tackling issues related to data scarcity and the absence of evaluation metrics. FinRpt features a dataset construction pipeline that amalgamates seven types of financial data, facilitating the automatic generation of a high-quality ERR dataset for training and evaluation purposes. Furthermore, an evaluation framework comprising 11 metrics is proposed to measure the quality of the generated ERRs. The authors also introduce FinRpt-Gen, a multi-agent system designed for this purpose, employing Supervised Fine-Tuning and Reinforcement Learning on the proposed datasets. This advancement represents a pivotal move toward the complete automation of equity research report production, with the potential to revolutionize financial analysis processes.