MT4 Modelling Quality: What 99 Percent Really Means

Written by Dominic Walsh · Published 12 Aug 2026 · Last updated 17 Aug 2026

Every MT4 backtest report opens with a percentage, and almost nobody reads it correctly. MT4 modelling quality does not score your strategy at all. It scores the simulation the platform built underneath your strategy.

This guide explains what MT4 modelling quality counts, why ninety percent is the practical ceiling on the platform’s own history, and where the famous ninety-nine percent figure comes from. It also covers the flaws that a high percentage leaves completely untouched.

The chart above shows the hourly bars a rule would trade on. Hold that window in mind, because a second chart later in this guide shows the same window in finer detail.

What MT4 Modelling Quality Actually Measures

The tester needs a price path inside every bar. Your rule may place a stop and a target that both sit within one hourly candle, so the simulation has to decide which came first.

The percentage describes how much of that inside-the-bar path came from stored data rather than from a construction. High means more real data. Low means more filling in.

It Says Nothing About Your Rules

A poor strategy on ninety-nine percent quality remains a poor strategy. The figure grades the raw material, not the recipe.

So treat it as a precondition. Only after the number looks sane does the rest of the report deserve any reading at all.

Where the Number Comes From

MT4 walks through your selected period and generates ticks for each bar. It scores each bar by how much genuine information supported that generation.

Bars with dense underlying data score well. Bars rebuilt from almost nothing drag the average down, and the report shows only the average.

The Field Reads n/a Sometimes

Choose Open prices only and the field shows n/a rather than a percentage. That makes sense, because the method builds no intrabar path to score.

Traders occasionally read n/a as a technical fault. It simply reflects the modelling choice you made when you started the run.

The Three Modelling Methods Behind the Figure

MT4 offers three ways to reconstruct the movement inside a bar. Each one produces a very different percentage and a very different report.

Open Prices Only

The fastest method evaluates your rules at the open of each bar and nowhere else. Nothing inside the bar exists as far as the simulation cares.

For a rule that decides at a bar open and exits at a later bar open, this can serve as a rough first pass. For anything with an intrabar stop, target or trailing step, it produces nonsense.

Control Points

Control points sketch the path using the nearest smaller timeframe. Test an hourly rule and the simulation leans on half-hourly bars, which is coarse.

The method runs fast, so it suits a quick check for coding errors. It does not suit any conclusion about profitability, because the path it draws skips most of the movement.

Every Tick

Every tick is the most precise option inside the platform. MT4 takes the one-minute history and interpolates a plausible tick sequence through each of those minutes.

Note the word interpolates. Unless you import real tick data with a third-party tool, the ticks remain a reconstruction rather than a recording.

The diagram above sets out what the percentage does and does not claim. Read it once and most of the confusion around the figure disappears.

Why Ninety Percent Is the Practical Ceiling

Run Every tick on MT4’s own one-minute history and roughly ninety percent is as high as the figure climbs. That number surprises traders who expected a hundred.

The Arithmetic Behind It

The platform holds one-minute bars, which means four prices per minute. Real trading inside that minute may involve dozens of quotes, and none of them survive in the record.

So the reconstruction fills the gaps with a rule of thumb. The scoring reflects that honestly, and it never awards full marks for a guess.

What a Low Figure Signals

A percentage well below ninety usually points at gaps in the stored bars. Whole days may hold no one-minute data at all.

Repair the history before you read anything else. A report built on half the period you selected describes a period you never chose to study.

A Common Trap on Fresh Installations

A new terminal downloads very little history by default. So the first Every tick run on a five-year range may score badly and cover far less ground than the dates suggest.

Our companion guide to the MT4 Strategy Tester covers the history preparation step in detail.

The Detail an Hourly Model Never Sees

Percentages stay abstract until you look at price. So compare two views of the identical stretch of market.

The chart above shows the same window as the first chart, this time at fifteen minutes. One window, two resolutions, and a very different amount of information.

Read the Two Charts Together

On the hourly view, a bar arrives as four numbers. On the finer view, that same hour resolves into several separate pushes, each with its own high and low.

Your stop lives inside those pushes. So does your target, and so does every trailing adjustment your rule tries to make.

Why the Order of Events Decides Results

Take a bar where price touched both your stop and your target. The outcome hinges entirely on which arrived first.

A coarse model has to pick, and its pick may be wrong. Repeat that across a few hundred trades and the equity curve tells a story the market never told.

The Same Problem in Multi-Timeframe Rules

Rules that read a higher timeframe while trading a lower one face this twice over. Our guide to multi-timeframe analysis explains how the two clocks interact.

A Worked Example Inside One Bar

Abstract talk about paths hides the point. So follow one trade through a single hourly candle.

The Setup

Your rule buys at the open of the hour. It sets a stop below the previous low and a target the same distance above entry.

During that hour, price dips to within a hair of the stop, then rallies through the target. A second dip later in the hour goes deeper still.

What Each Method Concludes

Open prices only sees the entry and nothing more. It waits for the next bar open, so both the near miss and the target simply do not exist.

Control points sketches two or three waypoints. It may catch the rally, and it may equally place the deep dip before it.

Every tick walks a fine path through the hour. It has a fair chance of ordering the events the way they truly happened.

Why the Difference Compounds

One trade turning from a loss into a win looks minor. Repeat the ambiguity across three hundred trades and the curve changes shape entirely.

Worse, the error is not random. Coarse models tend to favour whichever price the reconstruction reaches first, and that bias runs in one direction for a given rule.

Modelling Quality Beyond MT4

Other platforms handle this differently, and the vocabulary changes with them. Two points travel across all of them.

Every Simulator Guesses Something

MT5 offers its own tick modelling options and can use real ticks where a broker supplies them. Web-based backtesters usually work from bar data alone.

Names differ, and the underlying problem does not. Somewhere in every simulation, software decides an order of events that nobody recorded.

Read the Documentation, Not the Marketing

Find out what resolution a platform actually uses before trusting its numbers. A tool that never mentions intrabar handling has usually chosen the crude option.

Then apply the same questions you would apply to MT4. Data source, spread model, trade count and held-back data matter everywhere.

Where Ninety-Nine Percent Comes From

Vendors quote ninety-nine percent constantly. That figure does not come from MT4’s own history.

Imported Real Tick Data

Third-party tools pull recorded tick streams from data providers and load them into the tester. The simulation then reads quotes that genuinely printed instead of quotes it invented.

Scores climb sharply as a result. The path inside each bar now rests on evidence, so the scoring rewards it.

What Improves and What Does Not

Order of events improves enormously, and so does the treatment of fast moves. Variable spread also becomes possible, since many tick sets carry both sides of the quote.

The strategy underneath stays exactly as good or bad as before. Better data measures the rule more accurately, and it never upgrades the rule.

Whose Ticks Are They?

Imported ticks come from one provider, and your broker is a different company. Quotes differ, spreads differ, and the daily close hour may differ too.

So even a ninety-nine percent run describes a market adjacent to yours. Our deeper guide to tick data backtesting works through the trade-offs.

What a High Figure Still Cannot Fix

Traders treat ninety-nine percent as a stamp of approval. Five serious problems survive it untouched.

Curve Fitting

Precise data measures a fitted rule precisely. If you tuned nine inputs until the curve looked handsome, better ticks simply confirm the fitting in higher resolution.

A Sample That Is Too Small

Forty trades remain forty trades at any modelling quality. Chance produces streaks of that length routinely, so the sample cannot separate skill from luck.

Missing Costs

Commission and swap sit in the account specification, not in the tick stream. Leave them out and the report overstates every result, however fine the data.

Latency and Slippage Variation

No tick set contains the delay between your signal and your broker’s fill. Nor does it contain a requote or a partial fill.

The Trader

Hesitation, early exits and a hand on the off switch never appear anywhere. Yet those three end more automated records than any market condition.

Our list of common backtesting mistakes covers the rest of the pattern.

Mismatched Charts Errors and Gaps

Beside the percentage sits a count of mismatched charts errors. Traders skim past it, which is a mistake.

What the Count Means

It means the one-minute bars and the higher timeframe bars disagree. The tester rebuilt the larger candles from the smaller ones and found inconsistencies.

So the message flags a data problem, never a strategy result. Nothing about your rules caused it.

Why Gaps Appear

Records expire, feeds change and downloads stop early. Weekends and holidays add legitimate gaps that the checker still counts.

Rebuild the history, then rerun. A count in the thousands makes the rest of the report unreadable.

Weekends Are Not Faults

Some gaps belong there. Markets close, so a quiet stretch across a weekend or a public holiday reflects reality rather than a broken download.

Learn to tell the two apart before you start repairing anything. A missing Tuesday matters; a missing Sunday does not.

The Order of Operations

Fix data first, then settings, then rules. Reverse that order and you spend a week tuning a strategy against noise in a broken record.

How to Raise the Figure Honestly

Four steps move the percentage up without inventing anything. Work through them in order.

The panel above ranks the three levels by the detail each one carries. Detail rises as you move down the rows, and so does run time.

Download a Full One-Minute Record

Start with the deepest one-minute history your terminal will fetch for that symbol. Everything else builds on it.

Test a Range Your Data Actually Covers

Match the date range to the record you hold. A shorter honest test beats a long test with holes in the middle.

Choose Every Tick for Intrabar Rules

Any rule with a stop, a target or trailing logic needs the intrabar path. Control points and Open prices only both fail that requirement.

Import Real Ticks When Precision Decides the Outcome

Scalping rules and tight-stop rules justify the extra work. A weekly swing rule usually does not, since its stops sit far from the noise.

When the Figure Barely Matters

Modelling quality does not carry equal weight for every strategy. Two questions settle how much attention it deserves.

How Far Is the Stop From Entry?

A stop two hundred points away rarely gets touched by intrabar noise. So the reconstruction has little chance to mislead you.

A stop eight points away lives entirely inside the noise. Then the tick path decides almost every outcome.

How Long Do Trades Last?

Positions held for days depend on daily direction rather than the sequence within an hour. Positions held for minutes depend on almost nothing else.

So a swing rule can tolerate ninety percent comfortably. A scalping rule tested at that level tells you very little.

How Wide Is Your Spread Relative to the Move?

A rule aiming for ten points pays a large share of that target as cost. Small errors in the price path then swamp the result.

A rule aiming for three hundred points barely notices. So the same modelling quality carries different consequences for the two.

Symptoms of a Model That Is Too Coarse

Certain patterns in a report point straight at the modelling method. Four of them show up regularly.

Impossibly Clean Exits

Trades that close at the exact target price, over and over, suggest the simulation never tested the stop properly. Real fills scatter around the level.

Losing Clusters That Never Appear

A curve that rises with barely a pause deserves suspicion rather than admiration. Markets produce clusters of losses, and a model that hides them has smoothed something away.

Check the longest adverse sequence in the trade list. When that figure looks implausibly short, the intrabar path is usually the culprit.

Results That Swing With the Method

Run the same rule on Control points and on Every tick. A large gap between the two means the rule depends heavily on intrabar sequence.

That dependence is worth knowing early. It tells you precisely how much the data quality matters for this particular strategy.

Trade Counts That Change Between Runs

If the number of trades shifts when only the method changes, your entries themselves depend on the path. Then coarse modelling is not merely imprecise; it changes which trades happen at all.

Reading a Report That Quotes Ninety-Nine Percent

Vendor reports lead with the figure because it looks impressive. Four questions put it in proportion.

Which Data Provider Supplied the Ticks?

Ask for the source by name. Providers differ, and a report that cannot name its data source deserves no weight.

What Spread Model Ran?

Real tick sets allow variable spread, and plenty of reports still apply a flattering fixed number. Check which one produced the numbers.

How Many Trades and Over What Period?

Precision across sixty trades in one calm year proves nothing. Count the trades before you look at the curve.

Was Anything Held Back?

Ask whether the settings met a period the optimiser never saw. Without that step, the report describes a memory rather than a test.

Turn whatever trade statistics you do get into a single per-trade figure with our expectancy calculator.

A Practical Workflow

Put the pieces together and the routine becomes short. Six steps, in this order.

Step	What to do	What to check
1. History	Download the deepest one-minute record for the symbol	The record reaches the start of your intended range
2. Range	Set dates that the stored bars genuinely cover	No long empty stretch inside the period
3. Method	Select Every tick for anything acting inside a bar	The report shows a percentage rather than n/a
4. Score	Read modelling quality and the error count together	Around ninety percent, with a low error count
5. Costs	Set spread, commission and swap from your own account	All three appear in the account specification
6. Judge	Read the trade list, then the drawdown, then the curve	Enough trades across varied conditions to mean something

Then size the worst stretch against your real balance with our drawdown calculator. A percentage on a report feels abstract until it appears in account terms.

The Honest Summary

Modelling quality answers one question well. It tells you how much of the simulated path rests on data instead of construction.

It answers nothing else. A rule can score ninety-nine percent and still be fitted, undertested, mispriced and unfit for your account.

So use the figure as a gate rather than a verdict. Once it looks sane, the real work begins on the trade list, the costs and the unseen data. For the tools these rules usually start from, browse our MT4 indicators library.

FAQ

Is ninety percent modelling quality good enough?

For most swing and position rules, yes. Around ninety percent is the practical ceiling on MT4’s own one-minute history with Every tick, so it represents a normal, healthy run rather than a compromise. Rules with very tight stops or very short holding times need better than that, because the sequence of prices inside each bar decides most of their outcomes.

Why does my report show n/a instead of a percentage?

Because you selected Open prices only. That method evaluates your rules at the open of each bar and builds no path inside the bar, so no intrabar detail exists to score. Switch to Every tick if your rule places a stop, a target or a trailing step, since those all act between bar opens.

Does ninety-nine percent make a backtest trustworthy?

It makes the price path more faithful, and nothing more. Curve fitting, a thin sample, missing commission, absent slippage and your own behaviour all survive the upgrade untouched. Better data measures a rule more accurately; it never improves the rule itself.

What causes mismatched charts errors?

Gaps or inconsistencies in the stored history. The tester rebuilds higher timeframe candles from one-minute bars, and where the two disagree it raises the count. Treat it as a signal to repair or reimport the data, then run the test again before reading any performance figure.

Can I raise modelling quality without importing tick data?

Up to a point. Downloading a complete one-minute history, choosing a date range that the stored bars actually cover, and running Every tick will lift most poor scores back toward the platform ceiling. Beyond that, only recorded ticks from a data provider push the figure higher, and that step adds cost and complexity that many strategies do not need. Results are not guaranteed; past performance is not indicative of future results.