Backtest, Forward Test and Demo: What Each One Proves
What a backtest, a forward test and a demo account each can and cannot show, what to write down before a forward test, and how wide the noise is at 10, 30, 100 and 300 trades.
3 Nov 2026 · 6 min read
On this page
A backtest shows how a set of rules behaved on data that already existed. A forward test shows how they behave on data that did not exist when you wrote them. A demo account is the usual place to run a forward test, and it proves less than trading with money can. None of the three proves that the next trade will win.
What each one uses and shows
| Step | What it uses | What it can show | What it cannot show |
|---|---|---|---|
| Backtest | Historical data, replayed by you or a program. | How the rules would have behaved on that sample, quickly and on a lot of history. | Whether you chose the rules because of that sample. Real fills. How you will act through a losing run. |
| Forward test on a demo account | Live prices as they arrive, with rules fixed in advance and no money at risk. | Whether the result holds on data the rules never saw, and how the rules behave with a broker's real spread and platform. | Live fills: a demo account's execution is the broker's own simulation and can differ. The pressure of real money. |
| Trading with money | Real fills, real costs, real consequences. | What actually happens to your account. | Anything in advance: it is the most expensive way to find out the rules were weak. |
Write the plan before the forward test starts
A forward test only means something if the questions were fixed first. Otherwise you will read whatever number arrives as a pass. Decide, in writing:
- The rules, frozen. Changing one restarts the count.
- The number of trades you will take before you look at the result.
- What you will compare: the same figures you measured in the backtest, such as expectancy after costs, win rate and the deepest dip.
- What would count as the rules failing, and what would count as inconclusive.
- What you will do with each outcome, so the result decides it and not the mood you are in.
Why the number of trades matters so much
Even if the rules have a true average, a short run of trades will not measure it cleanly. The table uses made-up inputs: a rule that wins 45% of the time for 2.0R and loses the rest for 1.0R. Its true expectancy is 45% × 2.0 − 55% × 1.0 = +0.35R per trade, and one trade's result has a standard deviation of 1.49R. The standard error of an average over n trades is that figure divided by the square root of n.
| Trades | Standard error | Rough range of the measured average (true value ± 2 standard errors) |
|---|---|---|
| 10 | 0.47R | −0.59R to +1.29R |
| 30 | 0.27R | −0.19R to +0.89R |
| 100 | 0.15R | +0.05R to +0.65R |
| 300 | 0.09R | +0.18R to +0.52R |
After 30 trades, a rule whose true average is +0.35R could plausibly read anywhere from −0.19R to +0.89R, so a 30-trade forward test could not tell it from a rule that loses. This is the standard formula applied to made-up inputs, and it assumes each trade is independent of the last, which real trades are not always. Treat the table as the shape of the problem, not as a threshold. Its point is the second column shrinking slowly: ten times the trades narrows the range by a little over three times. The sample-size guide goes further.
Running the forward test on a demo account
- Treat it as practice for the process. Take every trade the rules give you, at the size you would use, and log it the same way as the backtest.
- Expect small differences in fills. Spread, slippage and execution on a demo account can differ from a live one at the same broker. Record them rather than ignoring them (see slippage).
- Do not tune as you go. If you must change a rule, start the count again and say so in the log.
- Compare like with like. If the backtest assumed a cost per trade, subtract the same one from the forward results, or the comparison starts out unfair.
Where a result from this procedure ends up is a decision for you. Nothing in a backtest or a forward test promises how a strategy will perform later. The worked sums here are arithmetic, not advice.
Trepeat today
Trepeat does not sell a backtesting or replay tool.
We're working on a free journal and a backtester, with no date yet. Today Trepeat is a trade copier for MT4, MT5 and cTrader, currently on demo accounts, and you can add notes, a grade and screenshots to any copied trade once it closes, with a calendar of your results.
What it supports now is the forward-testing part: rules you run on one demo account can be repeated onto others, and each closed trade can be written up in the journal. The backtest is the part you do elsewhere; the by-hand steps and a dated look at the tools are written for that.
Written by the Trepeat team.