Backtest vs forward test vs walk-forward: three tests, and only one runs in real time

All three test a strategy, but they answer different questions. A backtest asks how your rules would have done on the past; walk-forward, whether parameters chosen on one stretch of the past keep working on the next stretch, which they never saw; a forward test, whether the strategy holds up as prices arrive, when nobody knows how the candle ends. The first two still run on history, and only the third runs in real time.

Our explainers already cover each one on its own: backtesting, walk-forward and forward testing. This is about telling them apart, and about the places where the names themselves send you the wrong way.

The difference in one line each

• A backtest applies your rules to past prices, trade by trade. It tells you how the rules would have done in that stretch of history.
• Walk-forward optimises on a window of the past, tests on the stretch that follows, slides forward and repeats. Only the test stretches count. It tells you whether the way you choose parameters survives data it did not choose.
• A forward test runs the strategy going forward, on a demo account or live at minimum size. It tells you whether the edge survives an unknown future, and whether you follow the plan while it happens.

Side by side

Whether the outcome existed before the test is what sets the forward test apart: in a backtest and a walk-forward the market has already moved, however carefully the data is cut. What separates those two is the row below it, whether the data being scored could shape the parameters.

The trap in the name: MT5's "Forward" is not a forward test

The Strategy Tester in MetaTrader 5 (MT5) has a field called Forward, and its official help defines the term like this:

Forward testing is the repeated run of the best optimization results on a different time period.

In practice, the tester keeps the latest part of your dates out of the optimisation — a half, a third, a quarter, or everything from a start date you choose — and then reruns the best passes on it. That latest part is what the help calls the forward period.

That second part is still history. It had already happened when you pressed Start. What MT5 calls forward testing is a single out-of-sample cut: useful, and exactly the idea behind walk-forward, but done once instead of stretch after stretch.

It is not a forward test in the sense of our explainer, which runs on prices that did not exist yet. As of September 25, 2026, the same help page still does not mention walk-forward at all.

The consequence is practical: a report that says "forward-tested" can mean a date split inside the tester, or weeks of trading after the rules were frozen. Even a 2017 article on MQL5.com, the MetaTrader programming community, describes walk-forward as sticking together "forward test reports", for stretches that are all history. The words are the same, and what they prove is very different.

Why is none of them enough on its own?

The backtest can be tuned to the very period it scores. If the parameters were optimised on it, the result is the best of that period by construction, which is the mechanism behind a perfect backtest that fails live.

Walk-forward removes that possibility for the test stretches, as long as you do not then choose the window and step by how they came out, but it leaves the simulator's assumptions in place. In MT5, every stretch runs with the same tick model, a spread taken from history and, in its no-delay mode, every order filled at the requested price, which the platform itself calls "ideal" conditions. We described all three in what the MT5 Strategy Tester does.

The forward test is the only one the future can surprise, and it pays for that with time: a short sample says little. On a demo account the prices are live, but the fills are still simulated, so they may not show the slippage and rejections a real account meets — a similar limit to the one we pointed out in prop-firm evaluations. Live at minimum size adds them, along with the pressure of real money.

The order they go in

Backtest, then walk-forward or at least MT5's single forward cut, then a forward test on demo, then live at minimum size, and only then a larger size. Each step removes an assumption the previous one could not.

Measure every step with the same metrics. If they hold from one step to the next, the edge is surviving new conditions.

If they collapse, the previous step was assuming something that did not hold: between the historical steps that points to fit, and on the way to live it can also be execution or your own discipline. Either way, you found out before trading at full size.

The questions to ask when someone says a strategy is "tested"

• Which of the three? A backtest on its own says nothing about data outside its period.
• If it says "forward", which forward? A date split in the tester, or trading after the rules were fixed?
• If it says walk-forward, what were the window and the step, how many test stretches came out, and what did each one return?
• If it says live, on what kind of account? For robots sold on MQL5.com, the Market rules allow a link to a signal that monitors the product, and the Signals rules only accept real accounts; that is why a linked signal is the closest thing to a forward test you can see before paying, even if from the outside you cannot always check that it runs the same robot with the same settings.

What this does not settle

• The length of a forward test. It depends on how often the strategy trades, and we have no figure we can defend for all of them.
• Window and step sizes for walk-forward. Those belong to each strategy, and our walk-forward explainer does not propose them either.
• Any result of our own. We have not run these tests for this article; it compares methods, not strategies.

The takeaway

If you only keep one check, make it this: did the outcome exist before the test ran? If it did, you are looking at history, whether it is called a backtest, a walk-forward or MT5's forward period. Only a forward test answers with prices that had not been printed yet.