Walk-Forward Analysis
The biggest update OptiPie has seen so far. Backtests, stronger than ever with Walk-Forward Analysis.
Walk-Forward Analysis is here: optimization that has to prove itself on data it has never seen.
What is Walk-Forward Analysis?
Classic optimization runs a grid-search: it sweeps every parameter combination across your entire chart and keeps the best performer. The catch is built in. That “winner” was chosen because it fit the exact price history you tested it on. It is the top score on an exam the strategy already had the answer key to. Impressive in hindsight, and silent about what comes next.
Walk-Forward Analysis works differently. Instead of one pass over everything, it moves through your history in rolling windows:
- In-sample. It optimizes your parameters over a slice of history, exactly like a normal run.
- Out-of-sample. It then locks those winning parameters and tests them on the next slice of data, which the optimization never touched.
- Roll forward. The window slides ahead and the whole process repeats, window after window.
Every out-of-sample result is earned on unseen data. It is the closest thing to forward-testing without waiting months for it. That single change flips what the numbers mean: a grid-search tells you what would have worked, while walk-forward tells you what kept working once the market moved on.
Walk-Forward Efficiency (WFE) brings it together into one number: out-of-sample performance measured against in-sample. A healthy WFE means your edge held up on new data. A weak one means the backtest was fitting noise, and you learned it safely, before risking real capital.
See It in Action
A real walk-forward run, from the first window to the last.
Before the numbers, here is how a run is set up. Three settings shape every walk-forward analysis:
- Date range. The full span of history the analysis runs across, set by a start and an end date. This is the raw material every window is carved out of.
- Windows. Each window is one in-sample plus out-of-sample run. The date range is split into several of them, stepping forward through time, so more windows means smaller slices and more tests.
- Split. The in-sample to out-of-sample ratio inside each window. A 70/30 split optimizes on the first 70% of every window and tests on the remaining 30%.
Together they decide how much history each window sees and how often the strategy is put back on trial. The video below walks through a full run, a Supertrend strategy on BTCUSD split into 5 windows at 70/30, from setup all the way to the finished report.
In it, you will see the report OptiPie builds to be read at a glance:
- Summary metrics up top: average out-of-sample profit, how many windows stayed profitable, and the Walk-Forward Efficiency score.
- Rolling windows drawn as a staircase, so you can watch each in-sample and out-of-sample slice step forward through time.
- A per-window breakdown that puts each window’s best parameters and in-sample profit right next to what they actually made out-of-sample.
The walkthrough sets a run up from scratch, then reads the results window by window.
Ready to put your own strategy to the test? Open OptiPie and run a walk-forward pass on any optimization.