Comparing AI Stock-Selection Models Across Chinese Equity Universes

Summary

This weekly report compares machine-learning stock-selection portfolios across several Chinese equity universes: broad-market stocks without industry neutralization, broad-market selections neutralized to the CSI 300 or CSI 500 industries, and selections within the CSI 300 and CSI 500. It reports weekly, recent three-month, and recent one-year absolute and excess returns for approaches including random forests, stacking, logistic regression, naive Bayes, and support vector machines. The highlighted leaders vary by universe and measurement period; random forests lead several of the reported comparisons, while other methods lead some index-restricted results.

The report offers historical performance snapshots rather than a description of model inputs, training design, portfolio construction, or execution assumptions. Some reported portfolios had negative absolute returns despite beating their benchmarks, and benchmark-relative results also vary by period and universe. The source warns that models derived from historical experience can stop working. The figures are a dated comparison, not evidence that any model will continue to outperform or that results account for costs and implementation constraints.

Key ideas

• The report compares several AI model portfolios across broad-market, industry-neutral, and index-restricted Chinese stock universes.
• Random forests lead multiple reported categories, but the strongest model differs across universes and horizons.
• Performance is shown over weekly, three-month, and one-year periods using absolute and benchmark-relative returns.
• Some benchmark-beating portfolios still recorded negative absolute returns.
• The report warns that historical model relationships may fail and gives limited detail on methodology or implementation.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.