Machine Learning for Dynamic Multi-Factor Stock Rebalancing

Summary

The report describes a stock-selection strategy that predicts the future usefulness of seven style factors and adjusts their portfolio weights over time. It uses historical factor information coefficients (ICs), macroeconomic variables, and market variables as inputs to an XGBoost model. Factors with stronger predicted ICs receive more weight in a composite score, which ranks stocks for equal-weight selection from the highest-scoring group.

The report says rolling model training tracked changing market styles better than equal-weight factors or a fixed model in its historical backtest. It also notes that dynamic weighting increased turnover and that transaction costs eroded some returns, so limiting turnover improved the strategy. The reported performance figures are backtest results, not guarantees; the document warns that market structure, trading behavior, or greater participation in similar strategies could undermine the model. Some performance details in the source are incomplete, including the win rate.

Key ideas

• Predict future factor ICs using factor history, macroeconomic data, and market variables.
• Use predicted factor effectiveness to vary factor weights in a composite stock score.
• Select and equal-weight stocks in the highest-scoring group.
• Rolling model training is presented as a way to adapt to changing market styles.
• Higher turnover can consume returns through trading costs, and backtest results may not persist.

Tags

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