WFE thresholds: when is 0.7 PASS and 0.5 ACCEPTABLE in walk-forward testing
Practical guide to PASS vs ACCEPTABLE style bands for walk-forward efficiency, with context on when thresholds are informative versus noisy.
If you search walk-forward efficiency threshold, walk-forward efficiency 0.7 pass, or WFE passes needed for valid test, you are usually trying to convert a continuous metric into a decision. This page explains how to use WFE thresholds without fooling yourself.
First: what WFE is
If you are new to the metric, read Walk-Forward Efficiency (WFE) explained. Walk-forward efficiency summarizes how much of your in-sample edge carries into out-of-sample segments across a walk-forward process.
Why "0.7 PASS" is not a law of physics
Any fixed band (including 0.7 PASS style language) only makes sense when:
- You have enough walk-forward passes for stability.
- Costs and data are modeled honestly.
- You know whether parameters were re-optimized each window or held fixed.
- You are not comparing apples across different markets and timeframes.
If you only have two windows, thresholds are mostly noise.
What typically makes thresholds informative
- Enough windows so one lucky OOS segment can not dominate (Minimum windows).
- Enough trades per OOS segment for stable metrics (How many trades).
- Aligned definitions of IS and OOS PnL (In-sample vs out-of-sample).
How to interpret "ACCEPTABLE" vs "PASS"
Treat these labels as screening bands, not trading rules:
- PASS might mean "transfer looks strong under the stated process."
- ACCEPTABLE might mean "borderline transfer; investigate windows and costs."
Exact wording depends on product version. Always read the report footnotes and the methodology page.
When thresholds break: regime change and costs
Walk-forward analysis regime change can make yesterday's "PASS" irrelevant tomorrow. Walk-forward optimization cost (fees, slippage) can turn a good efficiency print into a bad live experience.
FAQ
Is walk-forward 3 windows minimum always true? Walk-forward 3 windows minimum is a common rule of thumb, not a theorem. Thin windows can still be unstable.
What about walk-forward efficiency 0.5? If your process is borderline, widen the sample, tighten costs, or simplify the strategy before you add leverage.