Machine Learning for Chinese Large- and Small-Cap Rotation

Summary

This research note examines shifts between large-cap and small-cap leadership in Chinese equities, using the CSI 300 and ChiNext indices as proxies. It argues that rotation can reflect macroeconomic conditions, industry trends, and market sentiment, and that earnings and valuation alone may not align with price moves in time. The proposed approach builds features across those three areas to estimate relative strength.

The note compares feature engineering methods for a logistic-regression rotation model. It reports that principal component analysis, identifying trend versus range-bound states, and screening features with a stepwise-regression approach improved the model in its tests. A 12-month rolling window is presented as the best tested balance of win rate and payoff, with reported backtest statistics including a roughly 62% win rate, 29% annualized return, and 38% maximum drawdown. These results are historical and depend on proxy selection and test conditions; the note says performance tended to be better when the proxies had lower historical correlation and a clearer market-cap distinction.

Key ideas

• The study models Chinese large-cap versus small-cap leadership as a changing relative-strength regime.
• It derives predictive features from macroeconomic, industry, and market-sentiment information.
• It reports improved model results from principal component analysis and state-aware feature selection.
• A 12-month rolling window performed best among the tested alternatives under the study’s assumptions.
• Backtest results varied with index proxy choice and included substantial drawdown.

Tags

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