Walk-Forward Analysis

A standard backtest is a single run on a single period. It is susceptible to overfitting — parameters that happen to work on historical data but have no predictive value going forward. Walk-forward analysis is designed to detect this.

By forcing the strategy to be tested on data that was never used for optimization, WFA gives a more honest picture of whether the strategy has real edge.

WFA splits data into three chronological windows: training (50%), validation (25%), test (25%). The strategy is optimized on training data. Performance is measured on validation data (to guide the optimization). Final reported performance is on the test window only.

If test performance is close to training performance, the strategy generalizes well. A large gap (train: Sharpe 2.1, test: Sharpe 0.4) indicates overfitting.

The WFA results table shows Sharpe Ratio, max drawdown, win rate, and total return for each window. Focus on the test column. Also look at the consistency of direction — if the test window is profitable even when train performance is not exceptional, the strategy is robust.

A strategy that shows positive Sharpe in test across multiple different date ranges (run WFA on different historical periods) is much more credible than one that passes a single test window.