Comparing AI Stock Selection Models Across Chinese Equity Universes

Summary

This weekly report compares stock selection portfolios built with several artificial intelligence methods across broad Chinese A-share and index-constituent universes. It separates results for unconstrained selection, industry-neutral portfolios, and selection within the CSI 300 or CSI 500, showing that the relative leader varies with the universe and evaluation period.

The report gives weekly, roughly three-month, and one-year absolute and benchmark-relative returns for methods including random forests, neural networks, naive Bayes, logistic regression, and support vector machines. Random forests lead several recent windows, while other methods lead some index-restricted or longer-period comparisons. The figures are historical snapshots from 2017–2018, and several reported values are incomplete or the date ranges are truncated in the source. The report provides no model construction details, risk-adjusted analysis, transaction costs, or out-of-sample validation, and explicitly cautions that AI-based selection may stop working.

Key ideas

• Model rankings differ across the broad-market, industry-neutral, and index-constituent universes.
• Random forests lead several reported recent-period comparisons, while other algorithms lead in selected settings.
• The report compares absolute returns and benchmark-relative returns over weekly, three-month, and annual windows.
• The results are historical and do not establish that any model will remain effective.
• Some figures and evaluation dates are incomplete in the source.

Tags

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