Reading a missing figure as a good figure
An omitted modelling quality is not a modest one; it is an unmeasured one. Any default filled in on the reader's behalf turns silence into a claim nobody made.
Modelling quality
A percentage the MT5 strategy tester prints for how completely the price history covered the tested window — reported on generated-tick runs too, not only on real ticks.
also: tick quality, history quality, backtest quality
Updated
In plain English
How closely the prices the tester used match the prices that actually happened. A low figure means the simulation invented most of the movement between the prices it knew, and the strategy traded the invention.
Why it matters
Every other backtest figure is computed on top of the price series the tester used, so modelling quality is the number that decides whether the rest of the report is measuring the strategy or the interpolation. It matters most exactly where EA marketing is loudest — intraday and scalping strategies, whose entire edge lives inside the bar.
- It bounds how much the report can be trusted. A profit factor computed on synthesised intra-bar movement is a fact about the synthesis.
- The damage is not uniform. A daily-close strategy barely notices the difference; a strategy with a 5-pip target can have most of its trades decided by prices the tester made up.
- It is the field most often left blank, and a blank field can be worse than a low one. The schema behind these listings defaulted it to 99, so a run done on generated ticks displayed '99% modelling quality, against a 92% minimum' until that default was removed on 2026-08-06.
Using it to validate a strategy
Where it is used
- Read at the top of every MT5 Strategy Tester report, printed as 'History Quality' with the modelling mode beside it.
- Quoted on an EA listing to establish what the published profit factor and drawdown were computed on.
- Chosen deliberately before a run: 'Every tick based on real ticks' (Model 4) for anything intraday, faster models only for coarse parameter scanning.
- Compared between two runs of the same EA to explain why their figures disagree.
What a reliable result looks like
- The tester reports real ticks and a history quality at or near 100%, and the data source is named.
- The instrument's tick history genuinely covers the whole window rather than being back-filled from bars for the early years.
- The spread in the data is the recorded spread rather than a fixed value standing in for it.
- One modelling mode covers every run in the comparison.
Warning signs
- No modelling quality stated at all — the absence used to read as a respectable default, which is why it must be measured rather than omitted.
- A high figure quoted without the data source, which is the half of the claim that can be checked.
- An intraday or scalping EA whose report shows interpolated bar-based modelling.
- A figure that changes between two runs of the same window, which usually means part of the history was filled in on one of them.
- A basket EA's figure judged by single-instrument standards. A run that loads several symbols at once reports a lower quality for that reason alone, so such a comparison ranks workloads rather than simulations.
Backtest vs live trading
- Interpolated data produces fills that were never available. The tester assumes a smooth path between known prices, so stops and targets trigger in an order the market never offered.
- Fixed-spread modelling removes the widening that live accounts pay, and the widening lands hardest on the small targets that low-quality modelling already mis-prices.
- Real-tick modelling narrows the gap but does not close it: latency, requotes and the broker's own execution remain outside the simulation.
- Expect the size of the backtest-to-live gap to track the modelling quality directly — the coarser the data, the wider the surprise.
Recommended validation method
- 1Choose the tester model before the run and record it with the result, rather than reporting whatever the last run happened to use.
- 2Use real ticks for anything that trades inside the day, and check first that the terminal actually holds any — the ticks.dat behind one run here was 123 KB, which is no inventory at all.
- 3Re-run the finished strategy on a second real-tick source and compare — agreement between sources is what makes the figure evidence.
- 4State the quality figure and the source together on every published run, so a reader can reproduce rather than trust.
Example
Four runs published on this site, all reading the same tester field. Three of them used the same model — Model 0, every tick generated from M1 bars — over the same broker history, and differed only in how many instruments the run had to load.
- Model 0, one instrument (GBPJPY)
- 99% history quality
- Single-symbol M1 history.
- Model 0, five instruments (basket)
- 72% history quality
- Model 0, seven instruments (basket)
- 72% history quality
- Model 4, real ticks (Dukascopy)
- 100% history quality
A basket run's own workload drags its figure down, so every basket listing here publishes no quality figure rather than a low one a reader would misread.
Calculation 99% → 72% with the model, the broker and the data source held constant; only the instrument count changed
Result The 27-point fall measures the run's history workload, not its fidelity
How to interpret it
Treat modelling quality as a qualifier attached to every other figure in the report, and read the strategy's holding time before deciding how much it matters.
| Range | What it means |
|---|---|
| Real ticks, quality at or near 100% | The figures are about the strategy. Remaining differences live in execution rather than in the price series. |
| Around 90%, bar-interpolated | Usable for end-of-day systems, misleading for anything that trades inside the bar. Read it as a scan, not a result. |
| Low figures used for parameter scanning | Legitimate as a search tool and worthless as a published claim. Any surviving candidate must be re-run properly. |
| Not stated | The weakest case, because it is indistinguishable from not having measured it. Ask before reading the profit figure. |
- Match the data to the strategy's timeframe. The finer the target, the more of the result was decided between the prices the tester knew.
- A high figure on an unnamed source is half a claim. The source is what makes it reproducible.
- When two runs of the same EA disagree, check the modelling mode before looking for a bug in the rules.
Of the 30 EA listings here, 6 state a modelling quality — all of them 100%, all on runs done on real ticks. The other 24 state nothing: 12 ran Model 0, where the tester builds every tick from M1 bars instead of replaying recorded ones — including the two basket runs whose reports print 72% — and 12 more real-tick runs report no figure at all. A build gate reads the shipped pages rather than the source, because the 99 default lived in the schema and a source-text check would miss the next way it comes back.
Common mistakes
Assuming a high percentage makes the backtest realistic
It only means the price series was faithful. Slippage, requotes, latency and the broker's execution policy all sit outside the tester, so a 100% run still flatters the fills.
Comparing figures produced under different tester models
The modelling mode changes the trade set, not just the prices — trades that trigger on interpolated data may never trigger on real ticks. Two such runs are different experiments.
Applying scalping standards to a daily system, or the reverse
A swing strategy holding for days barely feels the intra-bar path; a scalper lives on it. One threshold for both gets one of them wrong.
In depth
Frequently asked questions
- What modelling quality should an EA backtest have?
- For anything that opens and closes inside a day, real ticks with a history quality at or near 100%. For end-of-day systems, bar-interpolated data around 90% is usually adequate. The figure matters in proportion to how much of the strategy's outcome is decided between the prices the tester knew.
- Why is modelling quality below 100% even on real tick data?
- Because the tester reports how completely the history covered the requested window. Gaps in the broker's stored ticks, back-filled early years, a window that starts before the tick history begins, or a run that has to load several instruments at once all reduce the figure even when the run itself uses real ticks. The last of those makes a multi-symbol run's figure incomparable with a single-symbol run's.
- Is a 90% backtest useless?
- Not useless, but limited. It is a reasonable filter for slow strategies and a poor basis for publishing figures on a fast one, because interpolation supplies exactly the intra-bar path that a short-target strategy trades.