Walk-forward analysis

Walk-forward analysis is a backtesting methodology that repeatedly splits historical data into a training (in-sample) window and an immediately following test (out-of-sample) window. Parameters are optimized on the training window, applied unchanged to the test window, and the results recorded. The whole frame then rolls forward one step and the process repeats.

Why it matters. Standard backtests optimize and evaluate on the same data, making it trivially easy to overfit. Walk-forward forces every trade in the final equity curve to have been generated by parameters the model had never seen before — a much more honest measure of real-world performance.

Typical workflow:

1. Choose a training window (e.g., 180 days) and a test window (e.g., 30 days).
2. Optimize parameters on training; freeze them.
3. Simulate trades on the test window; record P&L.
4. Roll both windows forward by one test period and repeat.
5. Concatenate all out-of-sample periods into one equity curve.

A strategy with a strong out-of-sample Sharpe ratio after walk-forward is meaningfully more credible than one optimized end-to-end.

Read the full methodology in the walk-forward backtesting guide.