StockRanker: Ranking A-Shares with Price-Volume Factors and Gradient Boosting

Summary

This study presents StockRanker, a supervised learning-to-rank method for selecting Chinese A-shares. It uses gradient-boosted trees to rank the market by expected future return. The input features are derived from seven basic price and volume series using rolling averages, highs, lows, volatility, ranks, weighted averages, correlations, and lagged values. The baseline predicts five-day returns and buys highly ranked stocks, with portfolio weights tied to rank. The article compares a static training setup with periodically refreshed models and examines sensitivity to model settings and transaction costs.

Reported backtests show uneven results across years and parameter choices. In the static test, gains were modest relative to risk, while rolling retraining produced positive performance in many, but not all, annual periods. The study also shows that higher assumed trading costs can sharply reduce or erase returns, consistent with its high turnover. These are platform-reported historical simulations, not proof of future performance; the article itself points to data processing, model updates, and parameter selection as limitations and areas for further work.

Key ideas

• StockRanker applies gradient-boosted learning to rank stocks by predicted future returns.
• Its features are built from basic price and volume data using rolling statistical transformations and lags.
• The portfolio buys highly ranked shares and adjusts holdings as rankings change.
• Rolling retraining is intended to adapt the model to changing market conditions.
• Backtest outcomes vary across years and settings, and transaction costs materially affect results.

Tags

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