MASFIN: A Multi-Agent System for Decomposed Financial Reasoning and Forecasting
Marc S. Montalvo, Hamed Yaghoobian
TL;DR
MASFIN addresses the challenge of short-term stock forecasting in finance by integrating structured metrics, unstructured news sentiment, and explicit bias-mitigation protocols within a modular five-stage, multi-agent system. The approach combines open data sources (Finnhub and Yahoo Finance) with a HITL workflow to produce 15–30 stock portfolios while countering survivorship, hindsight, and overfitting biases. In an eight-week live evaluation, MASFIN delivered 7.33% cumulative return, outperforming major benchmarks in six weeks, albeit with higher volatility, and demonstrated strong alignment with market trends (SPY correlation ~0.97, QQQ ~0.95). The work contributes a transparent, reproducible framework that balances interpretability and performance, offering practical insights for deploying bias-aware generative AI in finance and other high-stakes domains.
Abstract
Recent advances in large language models (LLMs) are transforming data-intensive domains, with finance representing a high-stakes environment where transparent and reproducible analysis of heterogeneous signals is essential. Traditional quantitative methods remain vulnerable to survivorship bias, while many AI-driven approaches struggle with signal integration, reproducibility, and computational efficiency. We introduce MASFIN, a modular multi-agent framework that integrates LLMs with structured financial metrics and unstructured news, while embedding explicit bias-mitigation protocols. The system leverages GPT-4.1-nano for reproducability and cost-efficient inference and generates weekly portfolios of 15-30 equities with allocation weights optimized for short-term performance. In an eight-week evaluation, MASFIN delivered a 7.33% cumulative return, outperforming the S&P 500, NASDAQ-100, and Dow Jones benchmarks in six of eight weeks, albeit with higher volatility. These findings demonstrate the promise of bias-aware, generative AI frameworks for financial forecasting and highlight opportunities for modular multi-agent design to advance practical, transparent, and reproducible approaches in quantitative finance.