Glossary
Purged cross-validation
Cross-validation for time-dependent data that removes training samples whose labels overlap the test period, and adds an embargo after it.
Purging deletes from the training set every sample whose label window overlaps the test fold. An embargo also drops a short stretch of samples just after the test fold, because serial dependence carries information across the boundary. The method is described in López de Prado (2018).
It costs training data but removes the main source of inflated scores.
Example: 1,000 samples, a test fold of 200, labels spanning 10 observations and an embargo of 1% (10 samples) leave 781 training samples for a middle fold (9 purged, 10 embargoed).