Chinese Equity Selection Models: Weekly and Year-to-Date Results

Summary

This report summarizes stock-selection results for several machine-learning models across broad Chinese equity universes and constituent groups tied to the CSI 300, CSI 500, and CSI 800. It compares weekly, recent-month, and 2018-to-date excess returns, and also reports average RankIC for some model and universe combinations. The portfolios include all-share selections with industry and market-cap neutrality as well as selections within index constituents. Naive Bayes, random forest, XGBoost, SVM, stacking, and neural-network approaches appear in the comparisons.

The reported leaders vary by universe and measurement period: no single model dominates every comparison. The listed benchmark indices fell during the reported week, and some model portfolios also had negative absolute returns despite positive excess returns. The text is a summary of historical results, not a reproducible strategy specification; several values are missing in the supplied extract, and it warns that such models can stop working and may be difficult to interpret. It does not establish future performance.

Key ideas

• The report compares multiple machine-learning stock-selection models across Chinese equity universes and index constituent groups.
• It evaluates weekly, recent-month, and 2018-to-date excess returns, with RankIC averages reported for some comparisons.
• The model with the highest excess return differs across universes and time periods.
• Some strategies show positive excess returns while posting negative absolute returns during a falling market week.
• The report cautions that historical model strategies may fail and that their low interpretability calls for care.

Tags

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