Testing Boosting Models for Chinese Multi-Factor Stock Selection

Summary

This report compares AdaBoost, gradient boosting decision trees, and XGBoost as tools for selecting Chinese equities from factor data. Its workflow covers feature and label preparation, preprocessing, in-sample fitting, cross-validation, and out-of-sample testing. A rolling evaluation is used to adapt models to changing market characteristics, and the predicted probabilities of stocks rising in the next period are used to form sector-neutral portfolios drawn from major index constituents and the broader market.

The reported results show XGBoost outperforming linear regression on excess return and information ratio across the tested portfolio designs, though it does not show a clear maximum-drawdown advantage. Its out-of-sample AUC and accuracy are broadly similar to AdaBoost and GBDT, while training is reported to be faster. The report also describes its boosting trees as shallower and fewer than those in the random-forest comparison. These findings are specific to the study's Chinese equity universe, features, validation design, and backtests; the summary provides no detail on transaction costs or whether results generalize to other periods or markets.

Key ideas

• The study evaluates AdaBoost, GBDT, and XGBoost for multi-factor stock selection using rolling training and out-of-sample testing.
• Model predictions are converted into sector-neutral strategies for index constituents and the broader Chinese equity market.
• XGBoost is reported to outperform linear regression on excess return and information ratio, but not clearly on maximum drawdown.
• The three boosting models have similar reported predictive accuracy, while XGBoost trains faster in this comparison.
• The reported results depend on the study's data and backtest design and may not generalize to other markets or periods.

Tags

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