Purged Cross-Validation & Embargo Calculator

Ordinary time-series splits can still leak when labels use future observations. This tool makes the boundary mechanical: purge training rows whose label-information windows overlap validation, then embargo a further post-validation buffer.

Algorithmic TradingBacktest ValidationValidation leakage control

Part of the Grizzly Bulls backtest validation cluster. The parent guide explains how the checks and evidence benchmarks fit together; each child below owns a distinct quantitative task.

Validation checks

Execution friction Sequence risk Finite-sample win-rate uncertainty Research search multiplicity Serial dependence Sharpe estimation uncertainty Selection-adjusted Sharpe evidence Selection-process overfittingValidation leakage controlValidation path coverage

Evidence benchmarks

What can strategy search manufacture?What happens on untouched fresh data?How high should the winner clear?What changes when strategy variants are correlated?

Supporting concepts

Review the Financial Research Encyclopedia definitions behind important inputs, assumptions, or validation mechanics used on this page.

Embargo Period Label Overlap

100 rows, 5-row label horizonValidation rows 40–59 carry information through row 64.

5-row embargoThe reference case purges 10 training rows, embargoes 5 more, and retains 65 training rows.

Validation geometry

Total observationsForward label horizon (rows)Validation start indexValidation end indexEmbargo after validation information (rows)

Indices are zero-based and the validation window is inclusive. A label horizon of 5 means row i uses information through row i+5, capped at the sample end.

Leakage-aware split

20Validation rows

40–64Validation information interval

10Training rows removed for label overlap

5Additional post-validation embargo rows

65Retained training rows

81.3%Retained share of non-validation rows

Purging protects the label boundary

A row can sit outside the validation fold and still leak validation information if its target is computed from a forward window that reaches into the validation period. Purging removes every candidate training row whose label-information interval overlaps the union of the validation labels' information intervals.

For a training interval [t0, t1] and validation information interval [T0, T1], this implementation treats them as overlapping when t0 ≤ T1 and t1 ≥ T0. That is the mechanical leakage condition this calculator exposes.

Embargo adds a second buffer

Purging removes direct label overlap. Embargo removes an additional block of candidate training observations immediately after the validation information interval. The purpose is different: it creates extra temporal separation where nearby observations or features may remain serially dependent even after direct label overlap is gone.

The correct embargo is research-design dependent. This tool accepts an explicit row count rather than claiming one percentage is universally sufficient.

What this check does not prove

Purging and embargo reduce a specific form of validation leakage. They do not make observations IID, repair look-ahead features, fix survivorship bias, reconstruct missing point-in-time data, account for strategy search multiplicity, or create genuinely unseen future data.

Use the PBO / CSCV calculator to test the selection process across complementary splits, the autocorrelation check for serial dependence inside returns, and the Deflated Sharpe Ratio for selection-adjusted Sharpe evidence.

Reusable reference case

The checked-in reference case uses 100 observations, a 5-row forward label horizon, validation rows 40–59, and a 5-row embargo. It retains 65 training rows after removing 10 overlap rows and 5 embargo rows.

Download the reference case JSON

Method sources

Marcos López de Prado, Advances in Financial Machine Learning, Chapter 7, formalizes purging overlapping label intervals and embargoing observations after the test set.
Financial ML Core: PurgedKFold documents the interval-overlap rule used to remove training labels that intersect the test interval.