Using Macro Cycle Factors in Random Forest Stock Selection

Summary

The report combines cross-sectional equity factors with a set of three macro cycle indicators in a random forest model. The cycle features act as market state signals, allowing the model to apply different stock-selection logic across conditions. The authors suggest they can also emphasize more recent training observations during monotonic cycle phases and add information around turning points.

The report describes a six-month rolling training window and compares the augmented model with versions that omit cycle factors and with XGBoost. Its reported out-of-sample tests and China A-share, industry-neutral backtest show improvements in prediction metrics and portfolio performance over the historical evaluation period. The authors argue that random forests can use deeper trees to capture weak cycle-related signals more readily than shallow boosting trees. These are historical backtest findings, and the report warns that changes in market conditions could invalidate the learned relationships.

Key ideas

• Macro cycle indicators can help a stock-selection model adapt its factor logic to different market states.
• A six-month training window is proposed to respond more quickly to changing investment styles.
• The report argues that deeper random forest trees may capture weak macro signals more effectively than shallow XGBoost trees.
• Historical out-of-sample and portfolio tests report improvements after adding cycle factors.
• The observed relationships may fail if future market conditions change.

Tags

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