A technique for strategy evaluation that iteratively applies a moving in-sample/out-of-sample approach. In walk-forward testing, you divide historical data into multiple segments. For each segment, the strategy is first optimized on a training period, then immediately tested on the subsequent period that was not used in optimization. This process “walks forward” through the data (e.g., optimize on 2015–2017, test on 2018; then optimize 2015–2018, test on 2019; and so on). Walk-forward analysis simulates how a strategy might be continually re-optimized in real time. It helps ensure that the strategy’s parameters are not just curve-fitted to one static period. Essentially, “walk-forward backtesting” means the model is re-trained and re-evaluated at periodic intervals using new data. Consistent performance across multiple walk-forward segments is a sign of a robust strategy that adapts to changing market conditions.

Back to Robots Glossary Index