Time Series Cross-Validation Quant Interview Guide
Time series cross-validation quant interview guide for blocked folds, expanding windows, leakage, validation design, examples, and mistakes.
Candidates validating forecasting and signal models.
Time-series validation must respect order
Random folds can leak information when nearby observations are dependent or when future data informs past predictions. Time-series cross-validation keeps training earlier than evaluation.
Blocked folds reduce leakage
Blocked or chronological folds test later periods after earlier training periods. This better reflects forecasting and trading use cases than mixing all timestamps randomly.
Concrete example
Training on January through March and validating on April is more realistic than random days from all four months when the intended use is future prediction.
Expanding and rolling setups
An expanding window keeps adding data over time, while a rolling window uses a fixed recent history. The choice should match how the model would be retrained in practice.
Common mistakes
Candidates often reuse generic k-fold language. A stronger answer names the time ordering issue and designs validation around the actual decision timeline.
Practice the pattern
Use the LeetQuidity curriculum and calibration to turn this topic into a focused practice plan.