Tick Data Backtesting and Why It Matters

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

Every simulated trade needs a price, and every price needs a moment. Tick data backtesting replaces the platform’s invented path through each bar with quotes that genuinely printed, which changes how a strategy gets measured.

This guide explains what a tick actually is, where recorded tick sets come from, and what tick data backtesting fixes. It also covers the long list of problems that better data leaves exactly where it found them.

The panel above shows the whole idea in one picture. On the left sits a single bar. On the right sits the path price took to build it.

What Tick Data Backtesting Actually Changes

A standard MT4 run reconstructs movement inside each bar from one-minute history. It interpolates, which is a polite word for a reasonable guess.

Feed the tester recorded ticks instead and the guessing largely stops. The simulation now walks a path that a data provider observed rather than one the software drew.

The Order of Events

Most of the value sits here. When a bar touches both your stop and your target, the outcome depends entirely on which arrived first.

A guessed path picks one. A recorded path knows. Across three hundred trades, that difference reshapes the curve.

The Price You Get Filled At

Bar-based models fill you at neat levels. Recorded ticks scatter fills across the levels that actually traded, which looks messier and reads truer.

The Number on the Report

Modelling quality climbs sharply once recorded ticks arrive. Our companion guide to MT4 modelling quality explains why the platform’s own ceiling sits near ninety percent.

What Counts as a Tick

The word gets used loosely, so pin it down. A tick is a single update to the quoted price.

Not a Trade, a Quote

In spot forex there is no central exchange printing every transaction. What your terminal receives is a stream of quotes from your broker’s liquidity sources.

So a tick marks a change in the bid, the ask, or both. It does not certify that anyone traded at that price.

Two Sides, Not One

Each tick carries a bid and an ask. The distance between them is your spread at that instant, and it moves constantly.

Bar data usually stores one side only. That single simplification hides most of what a cost model needs.

How Many Arrive

An active pair produces thousands of ticks per hour during busy sessions. The same pair may produce a trickle at the Asian open.

Density itself carries information. Thin quoting periods behave differently, and a bar model erases the distinction entirely.

Where Recorded Tick Sets Come From

Nobody publishes one authoritative tick record for forex. Several sources exist, and they disagree with each other in small ways.

Data Vendors

Specialist providers record quotes from banks and brokers, clean them, and publish archives by symbol and year. Most third-party MT4 tick tools pull from these archives.

Your Own Broker

Some brokers expose a tick history through the platform or an API. That set matches your fills more closely than any vendor archive can.

Coverage tends to be short, though. Brokers rarely keep years of ticks available for download.

Recording It Yourself

This route costs nothing but patience, and it produces the closest match to your own fills. Start the logger on a machine that stays online, because every hour it misses becomes a hole later.

A small program can log every tick your terminal receives. Start today and in a year you hold a year of data, which does not help you test the last decade.

Why the Sources Disagree

Each provider aggregates a different set of counterparties. Prices diverge at the margins, especially during fast moves and thin hours.

So two ninety-nine percent runs on the same rule can produce different numbers. Neither is wrong, and neither is your broker.

Getting Tick Data Into a Test

The workflow is fiddly rather than difficult. Five stages cover it, whichever tool you use.

The diagram above lays out the sequence. Skipping the verification step at the end causes most of the confusion people report.

Choose the Period Deliberately

Longer is not automatically better. A decade of ticks describes markets with wider spreads and slower execution than the last two years.

Match the Symbol Specification

Point value, contract size, minimum volume and commission all belong to your account, not to the data set. Copy them across before the first run.

Verify Before You Believe

Run one short test, then check a handful of trades by hand against a normal chart. Errors at this stage look exactly like strategy results, which is how they survive.

One Window at Two Resolutions

Numbers stay abstract until price makes the point. So look at a single stretch of market twice.

The chart above shows an hourly window on GBPUSD. Each bar arrives as four numbers, and a bar-based test prices your orders against those four.

What a Bar Model Prices Against

Four numbers per hour is a thin description of an hour. Everything between them gets reconstructed by whatever rule the simulator applies.

For a rule with a wide stop, that reconstruction rarely matters. For a rule with a stop inside the noise, it decides nearly every outcome.

The comparison above summarises the difference in plain terms. Notice how much of the right-hand column concerns cost rather than price.

The Same Window, Closer In

The chart above shows the same window as the hourly view, now at fifteen minutes. One window, two resolutions, and far more movement on display.

Each hourly bar breaks into four, and each of those four carries its own high and low. A tick test prices against something finer still.

Why This Is Not a Trick of Scale

The extra swings are not decoration. Your stop, your target and every trailing adjustment sit inside them.

So the finer view is not a prettier picture of the same information. It is different information, and a coarse test never had access to it.

A Worked Example With One Stop and One Target

Theory about sequence stays vague until a trade hangs on it. So walk through one.

The Trade

Your rule buys at the open of an hour, with a stop twelve points below and a target twelve points above. Inside that hour, price first dips eleven points, then rallies thirteen, then falls twenty.

The correct answer is a winner. Price reached the target before it reached the stop, so the position closed in profit and the later fall never touched it.

What Each Model Concludes

A bar model sees a high, a low and a close. It knows both levels traded, and it has to choose an order without evidence.

Many simulators assume the adverse move came first. That assumption turns this winner into a loser, quietly and consistently.

A tick model reads the sequence. It closes the trade at the target, because that is what happened.

Why the Bias Runs One Way

Assumptions like this are not random noise. A pessimistic rule shifts every ambiguous bar in the same direction, and an optimistic rule shifts them all the other way.

So the error accumulates rather than cancelling out. Three hundred ambiguous bars produce a systematic distortion, not a wobble.

How to Spot the Exposure

Compare the trade list from two modelling levels on the identical period. Where the same rule produces different outcomes on the same bars, ambiguity is doing the work.

Count how many trades changed. That number tells you exactly how much your strategy depends on the intrabar path.

Cleaning and Checking an Archive

Raw archives arrive imperfect. Three checks catch the problems that matter before they contaminate a result.

Look for Missing Days

Plot the tick count per day across the whole period. Weekends should read near zero, and weekdays should not.

Any weekday reading zero marks a hole. A test that runs through that hole reports numbers for days the data never covered.

Look for Impossible Quotes

Bad records happen. A quote a hundred points away from its neighbours, or an ask below a bid, points at a fault in the archive rather than a market event.

Most conversion tools filter these automatically. Check what the filter removed, since an aggressive filter can delete genuine volatility along with the errors.

Compare Against Your Own Chart

Pick three dates at random and compare the archive against your broker’s chart for the same hours. Highs, lows and the timing of big moves should broadly agree.

Small differences are normal and expected. A gap of many points, or a move that appears on one and not the other, means the two describe different markets.

Variable Spread Changes the Cost Picture

Most standard runs apply one fixed spread to every fill. Recorded ticks carry both sides of the quote, so the test can charge what the market charged.

Where the Difference Shows Up

Spread widens around data releases, at the weekly open and through thin hours. Those moments happen to be when breakout rules and news rules trade most.

A fixed-spread run therefore understates cost precisely where the strategy is busiest. That is not a small rounding error.

The Rollover Hour

Liquidity thins as the trading day rolls over. Spreads there can multiply several times for a few minutes.

Any rule that trades that window looks better on fixed spread than it deserves. Our note on backtesting transaction costs covers the rest of the cost stack.

Compare What Your Own Account Charges

Vendor spreads and your broker’s spreads are two different things. Check the current picture with our forex spread comparison tool before assuming the test priced you correctly.

What Tick Data Cannot Fix

Better data measures a rule more accurately. It never improves the rule.

Curve Fitting

A strategy tuned across nine inputs until the curve looked handsome stays fitted at any resolution. Precise ticks simply describe the fitting in more detail.

Sample Size

Sixty trades remain sixty trades. Chance produces sequences of that length constantly, so no data upgrade turns a thin sample into evidence.

Latency

No archive contains the delay between your signal and your broker’s acknowledgement. Nor does it contain the moment your platform froze during a release.

Requotes and Rejections

A recorded quote stream shows what prices existed. It cannot show which of your orders a dealer would have refused.

Your Own Behaviour

Hesitation, early exits and a hand on the off switch appear in no data set. Those three end more automated records than any market condition. Our overview of what backtesting is makes the same point from the other direction.

Which Strategies Justify the Work

Tick testing costs time, storage and patience. Three questions decide whether it earns them.

How Far Is the Stop?

A stop three hundred points from entry sits far outside intrabar noise. A stop of six points lives inside it.

The tighter the stop, the more the tick path decides. That single ratio explains most of the answer.

How Long Does a Trade Last?

Positions held for a week depend on daily direction. Positions held for four minutes depend on almost nothing else but the tick sequence.

How Much Does the Rule Trade?

High frequency multiplies every cost error. A rule taking eight hundred trades a year pays the spread eight hundred times, so a wrong spread assumption compounds fast.

The Quick Test

Run the same rule on Every tick with platform history, then on recorded ticks. A large gap means your strategy depends on intrabar detail, and now you know it does.

The Practical Friction

Nobody mentions this part in the marketing. It matters more than the theory when you sit down to do it.

Storage and Time

Tick archives run to gigabytes per symbol per year. Runs that finished in minutes on bar data can take hours.

Optimisation multiplies that further. A sweep across two hundred combinations on tick data can occupy a machine overnight.

The Machine Matters

Tick runs lean hard on a single processor core and on disk speed. An older laptop can turn a twenty-minute job into an afternoon.

Plan around that before you commit to the approach. Nobody iterates carefully on a test that takes six hours to finish.

Setup Fragility

Tools that inject tick data into MT4 depend on folder paths, terminal versions and symbol names. A platform update can break the arrangement without warning.

The Temptation to Over-Trust

Effort creates attachment. After two days of importing and configuring, the resulting report feels far more authoritative than it should.

Guard against that. The work you invested in the data says nothing about the quality of the strategy.

Common Errors With Imported Ticks

Five mistakes turn a careful import into a misleading report. Each one has a plain fix.

Leaving the Default Spread in Place

Some tools import the ticks and still apply a fixed spread on top. Then all that work delivers a better price path with the old cost model bolted on.

Check the setting explicitly. If the report shows a constant spread column, the variable model never engaged.

Forgetting Commission

Commission lives in the account specification rather than the data. A raw-spread account charges it on every round turn, and a report without it overstates every result.

Mixing Symbols

Archive files carry provider naming, and your terminal carries broker naming. Load the wrong file and the test runs happily on the wrong instrument.

Confirm the first and last date of the imported set matches the symbol you intended. A silent mismatch here has ruined many afternoons.

Testing a Period You Also Tuned

Precision does not excuse fitting. Running an optimiser across recorded ticks produces a very accurate description of the past and no evidence about the future.

Trusting the First Run

Import problems and strategy results look identical on a report. Verify a handful of trades by hand before you accept anything the first run says.

Reading a Ninety-Nine Percent Claim

Sellers quote the figure because it sounds definitive. Four questions restore proportion.

Which Provider Supplied the Ticks?

Ask for the source by name and the years covered. A report that cannot answer deserves no weight at all.

Fixed Spread or Variable?

Plenty of high-quality runs still apply a flattering fixed number. Recorded ticks make variable spread possible; they do not make it automatic.

How Many Trades?

Precision across ninety trades in one calm year proves nothing. Count first, admire later.

Did Anything Stay Unseen?

Ask whether the settings ever met a period the optimiser never touched. Without that, the report describes a memory.

Convert whatever statistics you receive into one per-trade figure with our expectancy calculator. It cuts through a lot of presentation.

A Sensible Middle Path

Full tick testing suits a minority of strategies. Most traders get further with a staged approach.

Stage	Data used	Question it answers
1. Screening	Control points on platform history	Does the code run without obvious errors?
2. Shaping	Every tick on platform history	Does the rule survive a realistic intrabar path?
3. Held-back check	Every tick on a period never tuned	Do the chosen settings travel to unseen data?
4. Precision pass	Recorded ticks with variable spread	Does the edge survive real cost and real sequence?
5. Forward test	Demo on live prices	Does the rule behave the same outside a simulation?
6. Small live	Minimum position size	What do real fills and financing actually cost?

Stages one to three cost nothing but time. Reach stage four only for rules that survived the first three, since precision on a broken idea is wasted effort.

Where Slippage Enters

Stages five and six exist because no archive contains a fill. Our guide to slippage in trading explains what the gap between price and fill really costs.

The Honest Summary

Recorded ticks solve one problem properly. They replace a guessed intrabar path with an observed one, and they allow a spread that breathes.

Everything else stays where it was. Fitting, sample size, latency, requotes and your own behaviour all sit outside the data entirely.

So treat the upgrade as a measuring instrument rather than a verdict. For the rules that usually feed these tests, browse our MT5 indicators library and start from something you can describe in one sentence.

FAQ

Do I need tick data to backtest a strategy?

Not for most rules. A swing strategy with a wide stop and a holding time measured in days depends on daily direction far more than on the sequence inside an hour, so platform history on Every tick usually answers the question. Tight stops, short holding times and high trade counts are where recorded ticks start to earn their cost.

Why do two tick data sources give different results?

Because each provider aggregates quotes from a different set of counterparties. Prices diverge at the margins, particularly during fast moves and thin sessions, and those margins are exactly where tight stops get decided. Neither set is wrong, and neither one is your broker’s feed either.

Does recorded tick data include spread?

Usually yes, since a tick carries both a bid and an ask. That is what makes variable spread modelling possible, and it is arguably the biggest practical gain from the whole exercise. Check that your tool applies the spread from the data rather than overriding it with a fixed figure.

Can tick data show me slippage?

No. An archive records the prices that were quoted, not the price your broker would have given you, and it holds nothing about latency, requotes or partial fills. Those only appear once orders go to a live server, which is why a forward test remains a separate and necessary step.

Will a tick test match my live results?

Closer than a bar test, and still not exactly. The archive comes from a provider rather than your broker, so quotes differ at the margins, and no data set contains latency, requotes or the moments your platform lagged. Expect the tick run to narrow the gap rather than close it.

How much history should I import?

Enough to cover several distinct market conditions, without reaching so far back that the market stops resembling today’s. Two to five recent years usually strikes that balance for major pairs, and it keeps run times manageable. Weight what happened recently more heavily than depth of history, because spreads, execution speed and participants have all changed. Results are not guaranteed; past performance is not indicative of future results.