Why not use ordinary k-fold cross-validation for trading strategies?

Because it trains on data from after the test period and ignores overlapping returns, which leaks information and flatters results.

What is the difference between anchored and rolling walk-forward?

Anchored keeps all history in every training window; rolling uses a fixed-length window that moves forward.

What do purge and embargo do?

The purge drops training days whose outcome overlaps the test period; the embargo leaves an extra gap so nearby, correlated days do not leak.

Learn · Walk-forward validation

What is walk-forward validation?

Reviewed against the platform's code on Sep 23, 2026

Walk-forward validation tests an investment strategy in time order: it is fitted only on data before a cut-off, tested on the unseen period that follows, and the cut-off then moves forward and the process repeats — so every test result comes from a period the model had not seen.

Why it matters

Ordinary cross-validation shuffles time, letting a model learn from the future of the period it is tested on. In markets that inflates results, because returns overlap and regimes persist.

How it works

In an anchored (expanding) walk-forward, each training window starts at the beginning of the history and grows; in a rolling one it slides. A purge removes training days whose outcome window overlaps the test period, and an embargo adds a further gap.

Anchored walk-forward, five folds

Each row is a fold. Pale: the training window, which grows. Dashed: the purge and embargo gap. Solid: the test period.

How Opulence Alpha applies it

Opulence Alpha's strategy validation uses anchored walk-forward on the trading-day calendar, with a purge that covers the label horizon and an embargo at the train–test boundary, alongside purged combinatorial cross-validation.

Purged cross-validation →

Related concepts

Purged cross-validationPurged combinatorial cross-validation (CPCV) tests a strategy on many recombined train/test splits with overlapping periods purged, producing a distribution of outcomes instead of one lucky backtest path.Backtest overfittingBacktest overfitting is when a strategy is tuned to the noise of past data, so its backtest looks strong but its live performance decays. How to detect and avoid it.Point-in-time dataPoint-in-time data records what was actually known on each date — filing dates, revisions, delistings — so a backtest cannot use information from the future. Lookahead and survivorship bias explained.

All concepts →Sector behaviour in every market regime →

Questions

Why not use ordinary k-fold cross-validation for trading strategies?: Because it trains on data from after the test period and ignores overlapping returns, which leaks information and flatters results.
What is the difference between anchored and rolling walk-forward?: Anchored keeps all history in every training window; rolling uses a fixed-length window that moves forward.
What do purge and embargo do?: The purge drops training days whose outcome overlaps the test period; the embargo leaves an extra gap so nearby, correlated days do not leak.

References

López de Prado, M. (2018). Advances in Financial Machine Learning. Wiley.

Educational content about research methods. Not investment advice.