Comparing Ridge, XGBoost, and Random Forest Stock Strategies

Summary

This Chinese-language account describes a learning project that builds a China stock selection strategy with five ranked factors and a two-day forward price label. It trains a Ridge regression model, standardizes the inputs, and periodically retrains before ranking stocks for a small equal-weight portfolio. The author then explores backtest performance data, saves results, and compares Ridge, XGBoost, and random forest models using return, drawdown, Sharpe ratio, win rate, volatility, and other reported statistics.

The reported comparison favors Ridge on several listed metrics, while later tests show that results and factor importance can vary with model settings and feature combinations. The author emphasizes that factor selection, training frequency, and tree parameters are sensitive and that the exercise is for learning, not live trading. The document supplies no complete code for all models, independent validation, or evidence that the reported backtests account for overfitting or other sources of bias. Its discussion of low R-squared and IC is presented as informal guidance, not a demonstrated universal threshold.

Key ideas

• The example ranks five stock factors and uses Ridge regression to predict a forward price label.
• Stocks with the highest model scores are selected for an equal-weight portfolio, with periodic retraining and trading.
• The author compares reported backtest statistics for Ridge, XGBoost, and random forest models.
• Feature combinations and model parameters materially affect the reported results.
• The author cautions that the exercise is educational and not ready for live trading.

Tags

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