Walk-forward and out-of-sample testing for automated strategies

In-sample data is used to create or tune the strategy. Out-of-sample data is held back and examined only after the rules are largely fixed. Repeatedly checking the test set and changing the strategy to improve it gradually turns that data into another training set.

A clear research log should record when rules were frozen, which period was reserved, and whether any changes were made after seeing the result.

Forward testing runs the strategy on new incoming data, ideally in paper mode through the same alert and broker workflow intended for live trading. This reveals timing, symbol mapping, session, and order-lifecycle problems that a chart backtest may miss.

Compare the signal timestamps and theoretical prices with the actual paper orders. Differences should be explained rather than ignored.

Walk-forward analysis trains or calibrates on one historical window, tests on the next window, then moves both windows forward. It can show whether a strategy remains stable as market regimes change.

Be cautious when the process constantly retunes parameters. Frequent optimisation can add complexity and operational risk. A simpler rule that remains acceptable across windows may be more practical.