Guiding Random Forest Stock Selection with Preferred Investment Factors

Summary

This article describes a modification to random forest models that lets an investor express factor preferences directly in the tree-building process. Selected factors are forced to drive splits near the top of each decision tree, increasing their influence. The approach is motivated by the difficulty of adjusting complex machine-learning models as financial markets and factor effectiveness change. The modification adds controls for which features receive priority and how many upper tree levels use them.

The article applies the method to Chinese large- and mid-cap stocks, creating value, growth, and financial-quality portfolios with quarterly rebalancing. It compares settings for the depth of priority splits and reports that increasing this depth raises the selected factor’s importance in the model. This is a design proposal and portfolio test rather than evidence that the method reliably improves returns; the supplied text gives no detailed performance statistics or discussion of implementation risks such as overfitting or factor timing.

Key ideas

• The modified random forest prioritizes selected factors in the upper levels of its trees.
• The method is intended to make machine-learning models easier to steer as markets evolve.
• The article forms value, growth, and quality portfolios from the CSI 800 universe.
• Increasing the number of priority split levels raises the chosen factor’s model importance.
• The supplied summary does not establish improved investment performance.

Tags

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