What is walk-forward backtesting?
Walk-forward backtesting refits a strategy's weights at regular intervals using only data available at that moment, then measures what those weights earned in the period that followed. Every result after the first segment is out of sample, so the test cannot quietly reward a rule that was fitted to the answer.
How it works
A backtest that fits its weights once, over the whole history, and then reports how those weights would have done over that same history is grading its own homework. The weights already know which names worked. Nothing about the result tells you the rule would have helped anyone living forward through the period.
Walk-forward removes the loop. Refit on the first trading day of each new quarter or year using only what was knowable then, hold those weights through the segment that follows, record what they earned, and repeat. Stitching the segments together gives a track record where every period after the first was genuinely unseen when its weights were chosen.
The first segment is the exception and is not out of sample — it has no prior data to be fitted on. Read a short walk-forward accordingly: with three segments, one of them is a warm-up.
Refit cadence is a real trade. Refitting often keeps the weights current and pays more turnover and more estimation noise; refitting rarely is cheaper and slower to react. Neither is correct in general, which is why it is a control rather than a default nobody can see.
Where you see this in the terminal
Portfolio builder(account required)choose the refit cadence and the backtest becomes out-of-sample after the first segment
The allocators being testedwhat each weighting scheme uses, so you know what the refit is refitting
Reading the resultdrawdown conventions — the part of a backtest a return figure hides
The conventions behind it
What this page explains in plain language, the methodology documents state as conventions: the inputs, what is rejected, and the stated limits.
More of these: every concept page. The full transparency index is at Methodology.