XGBoost Fair-Value Modeling for Chinese A-Share Selection

Summary

This article adapts a machine-learning value strategy to Chinese A-shares. It trains an XGBoost regression model on accounting and company characteristics to forecast future price-to-book ratios. The gap between forecast and current price-to-book, scaled by recent volatility in that ratio, serves as a relative mispricing score for ranking stocks. The article describes winsorizing and cross-sectionally standardizing features, using an in-sample training and validation split, and testing on a later period. It also summarizes a related international approach that combines LASSO, XGBoost, and linear regression, and discusses profitability and news-sentiment filters from that research.

Key ideas

• The strategy treats predicted future price-to-book ratios as estimates of fair value.
• A standardized forecast-versus-current valuation gap ranks potential underpriced and overpriced stocks.
• The A-share implementation uses XGBoost regression with accounting and company characteristics.
• The described backtest rebalances annually and excludes the smallest stocks by market capitalization and trading value.
• The reported results cover a limited historical sample and do not establish that the approach will persist out of sample.

Tags

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