Walk-forward testing for short-term strategies

[Amanhasnoname]

A single optimization over a long period assumes the best parameters stay best. Markets change, and what worked in a quiet year might fail in a volatile one.

Walk-forward testing addresses this. Optimize on a window, say three months, then test on the following month without changes. Move the windows forward and repeat. The combined out-of-sample results show how the process performs under realistic conditions, including the need to re-optimize.

For scalpers, shorter windows often make sense because behavior changes with volatility regimes. A strategy that tunes itself monthly can track changes, but it can also chase noise.

The key output is not the profit figure but the consistency. If most windows are profitable and the parameters stay in a similar region, the approach has some robustness. If results swing wildly, it is fragile.

It is more work than a single test, but it also catches the most common failure: a strategy that only works on the data it was built on.

Has walk-forward testing been used, and how did the results compare with a single optimization?