Glossary · Backtesting & Analytics
Walk-Forward Optimization
Last updated:
Walk-forward optimization is a validation method that repeatedly tunes a strategy’s parameters on one window of historical data (in-sample) and then tests those frozen parameters on the following window (out-of-sample), rolling both windows forward through history. Instead of one optimization over the whole past — which invites curve-fitting — the strategy must keep proving itself on data it was never tuned on.
The output is a stitched-together equity curve made entirely of out-of-sample segments. If that curve still looks tradeable, the strategy’s edge is less likely to be an artefact of parameter fitting.
Example
Take six years of NIFTY futures data. Optimize a strategy’s stop and target on 2020–2021, then trade the best parameters on 2022 untouched. Roll forward: optimize on 2021–2022, test on 2023. Repeat until 2025. Suppose in-sample returns average 30% a year but the stitched out-of-sample curve manages 9% with deeper drawdowns. That 9% — not the 30% — is the honest estimate of the edge. If out-of-sample results collapse to zero, the “edge” was overfitting all along. Numbers are illustrative only.
A common split is 3:1 — for example, 12 months in-sample, 3 months out-of-sample, stepping forward 3 months each time.
Why it matters
Every optimizer will find parameters that made money in the past; that is arithmetic, not edge. Walk-forward analysis is how systematic traders separate the two before deployment, and it also reveals whether a strategy needs frequent re-tuning to survive — itself a warning sign.
In INDfolio AI, walk-forward style validation fits naturally after a first backtest on NSE data, before any capital is deployed.