Why it beats a single backtest
A single backtest tunes and reports on the same data. Walk-forward reports only on data the tuning never touched, so an edge that survives it is less likely to be luck.
glossary
Walk-forward analysis
The standard way to get an out-of-sample result from a strategy with parameters. OptAlgo runs it as one research run, with the same engine, fees and fills as the backtest.
Walk-forward analysis is a backtesting method that divides historical data into consecutive in-sample and out-of-sample windows. The strategy is optimised on each in-sample window and then run, unchanged, on the following out-of-sample window. The out-of-sample segments are stitched together into one equity curve, which estimates how the strategy would have performed if it had been re-tuned periodically on data available at the time.
Official Binance broker · Non-custodial · Live only when you confirm · Updated
Reading the verdict
OptAlgo returns robust, inconclusive or likely overfit, with the out-of-sample curve and the parameter path. A strategy whose best parameters jump wildly between windows is fragile even when the stitched curve looks fine.
Variants
Nested walk-forward adds an inner validation split; purged and combinatorial cross-validation handle overlapping trades; Monte Carlo and permutation tests check whether the result beats shuffled data. All are research runs on OptAlgo.
faq
Walk-forward FAQ
How many windows should I use?
Enough that each out-of-sample window holds dozens of trades. On 4-hour crypto strategies, yearly in-sample with quarterly out-of-sample windows is a common start.
Does walk-forward cost more backtest units?
Yes; it runs one backtest per window. Units are charged per 10 seconds of engine time across the run.