How to read MT5 backtest results: six numbers that decide whether an EA Trading Bot can be trusted
Every EA Trading Bot seller shows an equity curve that goes up. This article does not teach you to distrust the seller. It teaches you to read the report they hand you, so you can reach your own conclusion instead of simply believing or not believing.
A MetaTrader 5 Strategy Tester report has dozens of lines of numbers. Most of them do not matter. The six numbers below are enough to rule out most EA Trading Bots that are not usable, and they should be read in exactly this order.
1. Modelling quality: read it before anything else
This number tells you how closely the test simulates the market. MetaTrader calculates it from the quality of the price data you load. If you run on one-minute (M1) bars, it usually stops around 90%. If you run on real tick data, it reaches 99% or higher.
The gap between 90% and 99% is not cosmetic. The same EA can produce a beautiful equity curve on one-minute bars and lose money on real ticks. The reason is that at 90%, the software has to guess how price moved inside each bar, and that guess is always kinder than reality.
If modelling quality is below 99%, there is no need to read further yet. Not because the EA Trading Bot is certainly bad, but because that report is not yet fit to support any conclusion.
On the other hand, 99% is not automatically trustworthy either. If the historical data has gaps, or the spread is fixed at a level that does not exist in real trading, then 99% only measures the smoothness of something that was already wrong.
2. Maximum drawdown: more important than profit
Drawdown is the deepest fall of the account from its previous peak. This is the number to read before profit, because it decides whether you can hold on until the EA Trading Bot makes money.
The math is harsher than it feels. Lose 50% and you need a 100% gain just to break even. Lose 20% and you only need 25%. The difference between those two cases is much bigger than it looks.
| Drawdown | Gain needed to break even |
|---|---|
| 10% | 11% |
| 20% | 25% |
| 30% | 43% |
| 50% | 100% |
| 70% | 233% |
One more question matters just as much as the number: how long the drawdown lasted. A 20% drop that recovers in two weeks is very different from a 20% drop that stays at the bottom for eight months. The report does not tell you this; you have to look at the shape of the equity curve.
3. Profit factor, and the level where it becomes suspicious
Profit factor is gross profit divided by gross loss. Above 1 means the system made money. But the level worth paying attention to is much higher than that.
| Profit factor | How to read it |
|---|---|
| Below 1.2 | The edge is too thin; costs and slippage eat it all |
| 1.5 or higher | Acceptable |
| 2.0 or higher | Good, if the trade count is large enough |
| Above 5.0 over a short period | Most likely over-optimization |
The last row is the one people often read backwards. A nicer number does not mean a better EA Trading Bot. An unusually high profit factor over a short period is usually a sign that the parameters were tuned to fit the past too tightly, not that the strategy has a real edge.
4. Number of trades: with too small a sample, every number is meaningless
An EA Trading Bot that wins 9 out of 10 trades sounds impressive, but ten trades say nothing. That is a sample size pure luck can produce.
The commonly used threshold is 200 trades or more. Below that, any win rate and profit factor should be read as a hint, not as evidence.
5. Test period: it must include bad markets too
An EA Trading Bot that performs well during two years of uptrend has proved nothing about its ability to survive. The test period should be at least two years long, and more importantly it must contain every kind of market: trending, ranging, and at least one crash.
When you look at a report, ask for the start and end dates before you ask about profit. If the period was chosen carefully to avoid exactly the difficult months, the final figure tells you nothing.
6. Out-of-sample: the one number that is hard to fake
All five items above can be made to look good by tuning the parameters to fit past data. That practice has its own name: curve fitting, or over-optimization.
The only way to detect it is to split the data in two. One part is used for optimization and is called in-sample. The other part is kept completely hidden, never shown to the EA Trading Bot during optimization, and is called out-of-sample. A common split is to reserve one third to one quarter for the hidden part.
If results on the hidden part collapse compared with the part used for optimization, the strategy has no real edge; it is only memorizing the past. An optimization without an out-of-sample period is really just curve fitting with a few extra steps.
Sellers do not volunteer this part. Ask directly: which period was used for optimization, which period was kept hidden, and what the results were on the hidden period. The answer tells you more than any equity curve screenshot.
Net profit: always read it with the deposit and the time frame
Net profit is the end-of-period difference as the report accounts for it, but an amount of money cannot tell you the return or the level of risk on its own. Write down the initial deposit, the period and the drawdown before comparing configurations.
Expected payoff and recovery factor: an average and a ratio
Expected payoff divides the result into an average per trade; recovery factor puts net profit next to maximum drawdown. Both compress many trades into a single number, so you need to go back to the trade distribution and the length of the period to see what is hidden.
Win rate: not enough without the average win and loss
Win rate only counts how many trades closed in profit. Look at how much the average winning and losing trade made, along with open losing trades and equity drawdown.
A backtest is not proof
Reading all six numbers only gets you one third of the way. An EA worth running with real money needs to pass three stages.
- Backtest across several different periods, including trending, ranging and crash conditions.
- Forward test on a demo account with new data, for at least one to two months.
- Run it live with a small amount of capital before scaling up gradually.
A backtest can answer only one question: how this set of rules performed in the past. It cannot answer the question you actually care about, which is how it will perform tomorrow.
Frequently asked questions
It means the price data is detailed enough to simulate reality closely. But if the price history has gaps, or the spread is fixed at a level that does not exist in real trading, then 99% only measures the smoothness of something that was already wrong.
1.5 or higher is acceptable, and 2.0 or higher is good if the trade count is large enough. On the other hand, above 5.0 over a short period is usually a sign of over-optimization, not of a skilled EA Trading Bot.
There is no universal threshold, because it depends on each person's tolerance. But remember that losing 50% requires a 100% gain to break even, while losing 20% only requires 25%.
Usually 200 trades or more. Below that, win rate and profit factor should be read as hints rather than evidence, because pure luck can also produce good-looking results on a small sample.
It is tuning parameters to fit past data so tightly that the report looks good while the strategy has no real edge. To detect it, keep part of the data hidden from the optimization process, then compare results on the hidden part with the part used for optimization.
DinexSpace labels the source of every number in its research profiles: backtest, forward, demo or live. You can check them against the points covered in this article.