Walk-forward testing suits short-term strategies better than one big optimisation

[Amanhasnoname]

Optimising once on five years of data assumes the best parameters stay the best. Markets change, and settings that worked in one regime fail in another.

Walk-forward testing optimises on a window, tests on the next unseen period, then rolls forward. The out-of-sample segments combined give a more realistic picture of how the strategy performs when parameters must be chosen without knowing the future.

For scalpers, a window of a few months for optimisation and a few weeks for testing is typical. The results show how stable the parameters are. If the best settings change wildly from window to window, the strategy is fragile.

It also reveals decay. A strategy that works well for the first year and then fades shows up clearly in the walk-forward results.

The process takes more effort than a single test, but it gives an honest sense of robustness.

Has walk-forward analysis been used, and did the results differ much from a standard optimisation?