Model Drift Quant Interview Guide
Model drift quant interview guide for drift types, diagnostics, monitoring, retraining, regime shifts, examples, and tradeoffs.
Candidates discussing live model risk and changing market behavior.
Drift means the model environment changed
Model drift can come from changing inputs, changing relationships, market regimes, data pipelines, or user behavior. Name the suspected source.
Monitor inputs and outcomes
Track feature distributions, missingness, prediction distributions, realized performance, and known data-quality indicators. One metric rarely tells the whole story.
Concrete example
A volatility-sensitive model trained in calm markets may behave poorly during stress. Compare performance by regime rather than only averaging all periods.
Retraining is a tradeoff
Frequent retraining can adapt to changes but may chase noise. Slow retraining can be stable but stale. The right cadence needs evidence.
Common mistakes
Candidates often say retrain the model without diagnosing the drift. First identify whether the issue is data quality, distribution shift, or model decay.
Practice the pattern
Use the LeetQuidity curriculum and calibration to turn this topic into a focused practice plan.