FinRpt: Dataset, Evaluation System and LLM-based Multi-agent Framework for Equity Research Report Generation
Published 11 Nov 2025 · arXiv · Song Jin
Overview
The paper presents FinRpt, a novel benchmark for automating the generation of Equity Research Reports (ERR) using a multi-agent framework and large language models (LLMs). This addresses the challenges of data scarcity and lack of evaluation metrics in ERR generation.
Key Insights
- Dataset Construction Pipeline: Integrates seven financial data types to produce a high-quality ERR dataset for model training and evaluation.
- Evaluation System: Introduces 11 metrics to assess the quality of generated ERRs.
- Multi-agent Framework: FinRpt-Gen, a framework using LLMs trained with Supervised Fine-Tuning and Reinforcement Learning, demonstrates strong performance.
- Open Source: All code and datasets are publicly available, encouraging further research and development.
BFSI Relevance
- Why Relevant: Automating ERR generation can significantly enhance efficiency and accuracy in equity research, a critical component of financial services.
- Primary Sector: Financial Services
- Subsectors: Asset Management, Equity Research
- Actionable Implications:
- Financial analysts can leverage this framework to improve report generation efficiency.
- Asset managers can use these insights to enhance decision-making processes.
researcher peer-reviewed-paper global