That modelling quality figure in the MT4 strategy tester looks like a grade. It is not a grade of the strategy. It is a rough measure of how much of the test period was covered by real tick data, and the stock tester usually tops out around 90%. The 99% figure normally appears only after tick data from a third party gets imported.
Here's the mechanism. Brokers typically supply one-minute bars, not every tick. In "every tick" mode the tester fills the gaps between bars with generated ticks that follow a fixed pattern from open to high, low and close. Real price doesn't move like that. It stalls, spikes and reverses in ways that matter a lot to a scalper aiming for 3 pips.
The percentage also says nothing about spreads. Many tests run with one fixed spread for the whole period, while live spreads widen around rollover and news. A system that wins 60% of its trades at a fixed 0.8 pip spread can look very different at 1.8.
MT5 can test on real ticks collected from the broker's server, which is closer to reality, though history length and gaps depend on the broker.
So a beautiful report on tiny targets deserves suspicion. The smaller the target, the more a few tenths of a pip of tick error decides the result.
Have you ever run the same system on generated ticks and on real ticks, and how far apart did the profit factor land?
what the 99% in the strategy tester really means
-
Shadow Trader
- Posts: 5594
- Joined: Wed Sep 09, 2026 5:19 pm
-
FXS Prop Desk
- Posts: 1689
- Joined: Wed Oct 07, 2026 9:52 am
Re: what the 99% in the strategy tester really means
The 99% figure describes data coverage, not realism: it says nothing about variable spreads, slippage or execution delay. A test at 99% with a fixed 1 pip spread can still be less realistic than a 90% test with real spreads. Reporting the spread model next to the quality figure would make these numbers far more useful.
FXS Official Team | The Idea Lab & Strategy
-
Amanhasnoname
- Posts: 2060
- Joined: Fri Oct 30, 2026 9:45 am
Re: what the 99% in the strategy tester really means
I learned this the hard way with a little EURUSD grid scalper I bought years ago. The seller's report showed 99% modelling quality and a profit factor near 2.4, and I took that number as proof it was solid. When I imported Dukascopy ticks and ran it again with the real spread history, the profit factor dropped to about 1.1, and once I added a commission of seven dollars per lot it went below one.
The biggest difference came from the hour around rollover. On generated ticks with a fixed spread the robot looked brilliant in those quiet minutes, because it was picking off tiny reversions. With real spreads it was paying three or four pips to enter trades aimed at two.
These days I ignore the quality figure and ask two questions instead: what spread model was used, and how the result changes if I add half a pip to every trade. If the edge disappears at plus half a pip, I don't trade it.
The biggest difference came from the hour around rollover. On generated ticks with a fixed spread the robot looked brilliant in those quiet minutes, because it was picking off tiny reversions. With real spreads it was paying three or four pips to enter trades aimed at two.
These days I ignore the quality figure and ask two questions instead: what spread model was used, and how the result changes if I add half a pip to every trade. If the edge disappears at plus half a pip, I don't trade it.