Out-of-Sample Testing — what it is and how to use it in trading
Definition
Out-of-sample (OOS) testing splits historical data into two parts: a training part (in-sample), where the strategy's parameters are tuned, and a validation part (out-of-sample) — a separate stretch that took no part in the tuning, where the real result is checked.
How it works in the builder
In the backtest settings the OOS share is set by a parameter (typically the last 20-30% of history). If the OOS result is much worse than the in-sample one, the strategy is overfit to the past and is likely to perform worse going forward.
Typical use
A mandatory check before trusting backtest results: a strong result only on the in-sample part with no OOS check is a typical sign of a strategy tuned to fit one specific stretch of history. On the equity chart the OOS segment is usually highlighted separately, right after the training period ends.
Common pitfalls
A single-split OOS check is a weaker guarantee than walk-forward across multiple windows, since one test period can turn out to be atypical (for example, unusually calm or unusually volatile). Even so, it's the minimal, mandatory first step of an honest check.
This block is available in the strategy builder
137 no-code blocks — build an entry condition with this term and test it on history
Historical results do not guarantee future ones. The service is an informational and analytical tool, not an individual investment recommendation; trades are not executed.