08-12-2026, 08:51 PM
Optimising on all historical data and reporting the best setting is not validation. Walk-forward testing mimics the repeated research process.
Simple workflow
1. Choose a training window and a later test window.
2. Optimise parameters only on the training period.
3. Freeze them and evaluate once on the next period.
4. Move both windows forward and repeat.
5. Combine only the out-of-sample results.
Track parameter stability as well as returns. If a strategy needs a completely different setting every period, it may be fitting noise. Avoid repeatedly checking the test result and changing the rule—that turns the test set into training data.
What window lengths make sense for the market and timeframe you research?
Simple workflow
1. Choose a training window and a later test window.
2. Optimise parameters only on the training period.
3. Freeze them and evaluate once on the next period.
4. Move both windows forward and repeat.
5. Combine only the out-of-sample results.
Track parameter stability as well as returns. If a strategy needs a completely different setting every period, it may be fitting noise. Avoid repeatedly checking the test result and changing the rule—that turns the test set into training data.
What window lengths make sense for the market and timeframe you research?
Building small AI tools and automations in Bengaluru. Notes on what ships, what fails, and what I'm learning along the way.