Walk-forward Analysis
Pronunciation: wawk-FOR-ward a-NAL-uh-sis
Repeatedly optimising on one slice of history, then testing on the next unseen slice — rolling forward.
Definition
Walk-forward analysis splits history into a series of in-sample (optimisation) windows each followed by an out-of-sample (test) window, then "walks" both windows forward across the data. The strategy is re-optimised on each in-sample window and evaluated only on the following out-of-sample window, so the combined out-of-sample results approximate live performance far better than a single static backtest. It reduces — but does not eliminate — overfitting, and assumes the future resembles the recent past.
In plain English — Walk-forward analysis is a more realistic backtest: instead of tuning once on all of history, you tune on a window, test on the next unseen window, then roll both windows forward and repeat. Because every test happens on out-of-sample data, it mimics how you would actually re-tune and trade over time.
Example
Optimise on Jan–Jun, test on Jul; then optimise on Feb–Jul, test on Aug; and so on. You judge the strategy on the stitched-together out-of-sample months, not on any single tuned period.
Related terms
Where you see this in the app
Educational content only. Map.Trade does not provide financial advice or trading signals.