Out-of-sample (OOS) data is the portion of the historical record that the strategy is evaluated on after parameters have already been chosen. Because the strategy never saw OOS bars during tuning, its performance there is a much more honest estimate of how it will behave going forward than its in-sample performance.
The OOS / in-sample split is the single most important discipline in backtest design. Without it, every reported statistic is suspect — the developer has had countless chances to peek, re-tune, and try again until the numbers look good.
OOS discipline collapses if the developer keeps tuning until OOS results improve — at that point OOS has effectively become in-sample. The cure is to commit the strategy code first, run on OOS exactly once, and accept the result.
Comment Noon Barbari utilise Out-of-sample
Chaque concept ici est implémenté dans la plateforme. Ouvre la documentation ou l'outil concerné pour le voir en action.
OOS in walk-forward →Termes liés
- Backtesting
In-sample
The portion of history used to fit the strategy's parameters.
- Backtesting
Walk-forward optimization
Optimize on a rolling in-sample window, validate on the next out-of-sample slice.
- Backtesting
Overfitting
Fitting a strategy so closely to past data that it captures noise, not signal.
- Backtesting
Backtest
Simulating a trading rule on historical data to estimate how it would have performed.