Value Investing Rules Enhance Modern AI for Equity Selection

Modern finance often employs complex machine learning models to identify stock market patterns, but these models can sometimes overfit to short-term noise rather than identifying companies with enduring value. This study explored whether Benjamin Graham's foundational value investing principles could serve as a "low-pass filter" to stabilize and improve these advanced AI models. Researchers constructed three distinct feature sets: pure Graham rules, modern market factors, and a combination of both. These were then tested with sophisticated models like XGBoost and AutoGluon using two decades of S&P 500 data. A strict four-year buy-and-hold strategy (March 2022 to March 2026) was used for evaluation. The findings revealed that while complex algorithms like AutoGluon achieved high returns, they also incurred substantial drawdowns due to investments in volatile tech stocks before market crashes. In contrast, a Random Forest model incorporating pure Graham rules yielded the highest overall return with significantly lower risk. A combined Random Forest model also performed well, balancing momentum with Graham's rules to achieve strong returns while maintaining the lowest maximum drawdown. This research underscores that Graham's "margin of safety" remains highly relevant, effectively mitigating excessive risk in contemporary AI-driven investment strategies.