Walk-forward testing splits history into consecutive windows. You optimize parameters on one window, then test them on the next unseen window, and roll forward. The combined out-of-sample results are far closer to what you would have experienced in real time.
Choosing window sizes
A common starting point is an in-sample window three to four times longer than the out-of-sample window. Too short and parameters jump around; too long and the strategy can’t adapt.
Reading the results
Look for stability. If optimal parameters change wildly from window to window, the edge is probably noise. If they stay in a tight range, that’s an encouraging sign — though never a guarantee.
Important: Backtested results are hypothetical and are prepared with the benefit of hindsight. They do not represent actual trading and do not guarantee future performance. DataBacktest does not provide investment advice. See our Risk Disclaimer.
Have a dataset or a strategy you want tested properly?
Tell us what you’re working on. Mark or a member of the team will reply within one business day with a scope and a fixed quote.