Walk-forward analysis

Walk-forward analysis repeatedly tunes a strategy on one window and tests it on the next unseen window, exposing overfitting that a single backtest hides.

A single backtest optimised over the whole history is easy to overfit: with enough parameters you can fit the past perfectly and the future not at all. Walk-forward analysis guards against this by sliding through time — fit (or choose) parameters on an in-sample window, then measure performance on the following out-of-sample window the strategy has never seen, and repeat.

If the out-of-sample results hold up, the edge is more likely to be real. If they collapse the moment the strategy meets unseen data, you have curve-fitting dressed as a strategy.

Definitions are educational and consistent with Thuztra’s backtest methodology. Backtests are research, not investment advice; past performance does not predict future results.