Tag
This project open-sources an AI research system based on the frameworks of value investing masters like Buffett, Munger, Duan Yongping, and Li Lu. It uses Claude Code/Codex to enable multi-agent parallel analysis of financial statements, valuations, etc., and shows real trading returns of over 1.46 million yuan in two years, significantly outperforming major indices.
Introducing AI Berkshire – a collection of investment research Skills compatible with Claude Code and Codex, systematizing the methodologies of masters like Buffett, Munger, Duan Yongping, enabling professional-grade US stock research via AI Agent, and demonstrating real portfolio performance.
This research tests whether Benjamin Graham's classical value investing rules can act as a mathematical 'low-pass filter' to prevent modern machine learning models (XGBoost, AutoGluon) from overfitting to market noise. Using 20 years of S&P 500 data, the authors find that Graham's rules combined with Random Forest achieve high returns with lower risk than complex AI models alone.
AI Berkshire is a collection of investment research Skills based on Claude Code, systematizing the methodologies of value investing masters such as Buffett and Munger, enabling professional-level investment research through AI agents, with real trading performance records.