Walk-forward testing in plain terms
Posted: Fri Oct 30, 2026 1:54 pm
Optimising a strategy on all available history and then trading it live asks for disappointment. Walk-forward testing offers a more realistic check.
The method splits history into segments. The strategy is optimised on the first segment, then tested on the next segment without changes. The process then slides forward: optimise on a new window, test on the following one, and so on.
The combined out-of-sample results show how the optimisation process performs when facing the unknown, which resembles real trading far better than a single in-sample fit.
If out-of-sample performance is much worse than in-sample, overfitting is likely.
Several platforms support it, and basic versions can be done manually by dividing data into chunks.
It takes more effort than a simple backtest, but it screens out a large share of strategies that look brilliant only because of curve fitting.
Has walk-forward testing been used here, and how did the results compare with the single-pass backtest?
The method splits history into segments. The strategy is optimised on the first segment, then tested on the next segment without changes. The process then slides forward: optimise on a new window, test on the following one, and so on.
The combined out-of-sample results show how the optimisation process performs when facing the unknown, which resembles real trading far better than a single in-sample fit.
If out-of-sample performance is much worse than in-sample, overfitting is likely.
Several platforms support it, and basic versions can be done manually by dividing data into chunks.
It takes more effort than a simple backtest, but it screens out a large share of strategies that look brilliant only because of curve fitting.
Has walk-forward testing been used here, and how did the results compare with the single-pass backtest?