Comparing Linear Regression, XGBoost, and Random Forest on Four Stock Factors

Summary

This Chinese-language note describes a stock prediction exercise using four cross-sectional inputs: total market capitalization, turnover, trailing price-to-earnings, and quarterly net-profit year-over-year growth. It compares linear regression, XGBoost, and random forest as regression models, and reports that linear regression performed best in the author’s comparison. The note also says market capitalization and turnover had the largest influence among the selected factors, with returns falling noticeably when those two inputs were removed.

The author frames factor quality and model choice as complementary: factors supply the information available to learn from, while the model determines how that information is used. The comparison also highlights practical differences in training time, parameter needs, and interpretability. No model parameter search was performed; the note only mentions reducing the tree models’ estimator count to save time. It does not provide numerical performance, sample dates, validation design, or enough detail to assess whether the reported ranking generalizes beyond this exercise.

Key ideas

• The exercise predicts stock outcomes from four cross-sectional financial and trading factors.
• It compares linear regression, XGBoost, and random forest, reporting the strongest result for linear regression.
• The author reports that removing market capitalization and turnover materially reduced returns.
• Model choice affects training time, parameter requirements, and interpretability.
• The comparison lacks parameter optimization and does not provide enough methodological detail to establish general performance.

Tags

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