Forex glossary
Walk-forward analysis
Walk-forward analysis repeatedly selects parameters on an earlier window and tests them on the following period, then advances the split through time. It combines out-of-sample results under explicit cost and reoptimization rules. It evaluates the procedure without eliminating overfitting or constituting a live-account test.
Illustrative example
How does it work in practice?
Assume parameters are selected on January–March data and tested in April, then selected on February–April and tested in May. Each test uses previously chosen parameters; the April and May results are combined.
A case to work through
The test becomes part of selection
Hypothetically, parameters are selected on bars 1–100 and evaluated on 101–120.
- Viewing 101–120 and revising rules to improve it reuses test data.
- The new version needs a later evaluation window under the written procedure.
What follows? Do not present an improved result on that same window as an untouched test.
Example boundary: Repeated revisions can fit the whole procedure; walk-forward does not automatically remove this.
Chronological separation
Text description of the diagram
- Earlier window: Select parameters under the written procedure.
- Later test: Freeze parameters and assess subsequent data.
- Advance windows: Repeat without using the future test for selection.
A useful distinction
Walk-forward analysis / Backtesting
A backtest can assess one historical sample; walk-forward advances selection and subsequent evaluation windows.
Check your understanding
Recognize reuse of out-of-sample data.
Show the answer and explanation
51–60
Every bar follows the selection window.
From a plan to automation tests
Step 10 / 12 · 0 / 12 reviewed
View the full sequence →Progress is saved only in this browser.
Read the fuller explanation
Related terms
Sources and documentation
- QuantConnect – walk-forward
Periodic parameter selection over trailing historical windows and optimization-frequency assumptions.
Document read: - scikit-learn – time-series splits
Chronological train/test splits that place test observations after training data.
Document read: - MetaTrader 5 – optimization and forward testing
Historical optimization and a separate subsequent forward-testing segment.
Document read:
The example illustrates the mechanism. Check current account conditions and instrument specifications with the provider.