Lagged Features Quant Interview Guide
Lagged features quant interview guide for historical inputs, timing discipline, leakage prevention, validation, examples, and common mistakes.
Candidates designing features from historical observations.
Lagged features use past information
A lagged feature is built from observations available before the prediction or decision time. The key interview habit is to state exactly which timestamp each input uses.
Feature timing is the main risk
A feature can look lagged but still leak if the data was revised later, published with delay, or summarized using a window that includes the target period.
Concrete example
Using yesterday close to predict today may be valid if yesterday close was known. Using today high to predict today close is not valid for a decision made before the high occurred.
Validate lag choices
Different lags can be tried, but repeated lag search creates data-snooping risk. Keep validation chronological and be explicit about how many alternatives were explored.
Common mistakes
Candidates often say past data is safe without checking publication timing or window endpoints. In time-series interviews, past must mean known at decision time.
Practice the pattern
Use the LeetQuidity curriculum and calibration to turn this topic into a focused practice plan.