Comparative Study on the Investment Performance of Traditional Quantitative Methods and Machine Learning in Value Investing*

Abstract This study compares a quarterly‐rebalanced Piotroski F ‐score strategy with machine‐learning ranking models for stock selection in Taiwan's equity market using an expanding‐window walk‐forward design. Over the out‐of‐sample period 2021Q1–2023Q4, the Random Forest top‐20 portfolio achieves an annualized gross return of 27.3% (net: 23.1%) and a Sharpe ratio of 1.65, with a Carhart four‐factor alpha of 1.9% per quarter (). The F ‐score strategy also outperforms the benchmark. Sector‐neutral portfolios confirm that outperformance reflects within‐industry stock selection rather than sector allocation. Sensitivity analysis across portfolio sizes confirms robustness, and an extended 20‐quarter out‐of‐sample window preserves the main conclusions. Results demonstrate that ML enhances fundamental value strategies under realistic constraints.