Walk-Forward Analysis
A testing method that optimises parameters on one segment of history and evaluates them on the immediately following unseen segment, repeatedly.
How it is identified
Test: for each window, optimise on the in-sample period and record results only from the subsequent out-of-sample period
Unit
qualitative
In depth
Walk-forward analysis exists to detect curve fitting: parameters tuned on data the strategy has already seen will look excellent, and the honest measure is how they perform on data that came afterwards. Only the out-of-sample segments are counted, which typically reduces reported performance substantially and sometimes eliminates it. It also simulates the real process of periodic re-optimisation, so it tests the whole procedure rather than a single fixed rule. A strategy whose out-of-sample results collapse was never a strategy, only a description of the past.
Worked example
Optimising on 2018 to 2020 and testing on 2021 gives 6% against 31% in-sample. Repeating for four such windows produces an average out-of-sample result of 5% — the honest expectation, against a fitted 30%.
Illustrative figures, chosen so the arithmetic is easy to follow. Not a live price and not a valuation of any company.
Educational reference only
This entry explains what “Walk-Forward Analysis” means. It is not investment advice and not a recommendation to buy or sell any security. Any numbers above are illustrative, not live prices, and nothing here predicts price direction or rates a stock. Consider your own circumstances and consult a SEBI-registered investment adviser before acting.