Learn / Research process
Answer page / research process
Topic cluster / Research workflowWalk-forward testing repeatedly fits or chooses a strategy on one historical window, evaluates it on the next unseen window, and rolls the process forward so you can see whether the edge survives across multiple market conditions.
What to remember
Walk-forward testing is a repeated out-of-sample process. You choose parameters or strategy logic using one chunk of history, then test it on the next chunk that was not used to make the decision, and keep rolling both windows forward.
That is much closer to the real research problem than one polished train-test split because it forces the idea to survive several regime changes instead of one curated reveal.
A single out-of-sample slice can still flatter a weak strategy if that slice happened to be unusually friendly. Walk-forward testing asks whether the edge keeps reappearing when the market keeps changing and your process has to keep making fresh decisions.
The goal is not to optimize forever. The goal is to keep the research loop honest. That means realistic costs, fixed rules for when you are allowed to retune, and a clear distinction between model selection and evaluation.
Alphora's current public stack already leans toward this worldview: out-of-sample discipline, verified runs, and paper tracking as separate proof layers. The planned walk-forward sample sizer exists for exactly this reason. It is meant to help teams decide when they actually have enough evidence to stop flattering themselves.