Retraining Multi-Factor Models Before Live Trading

Summary

The document poses a model validation question rather than giving a complete answer: after training a multi-factor model on historical data and evaluating it on a later year, should a trader deploy that fitted model or retrain it on a newer rolling window? It also asks whether a newly trained model can be expected to reproduce the earlier model’s backtest performance.

The central issue is that changing the training period changes the observations available to the model and may change its fitted parameters and behavior. A satisfactory backtest on one held-out period does not ensure the same results after retraining or in live markets. The text contains no recommended retraining schedule, validation procedure, or empirical comparison, so it is useful chiefly for identifying model stability and out-of-sample generalization as questions that require further analysis.

Key ideas

• The post asks whether to deploy a previously fitted model or retrain it on a newer rolling window.
• Changing the training window can change fitted parameters and model behavior.
• A good backtest on one later period does not guarantee similar results after retraining.
• The text raises model stability and out-of-sample generalization but provides no procedure or evidence to resolve them.

Tags

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