Backtest Your EA in the Strategy Tester

What You Need Before You Open the Strategy Tester

A backtest is a controlled rerun of your Expert Advisor's rules over price history that has already happened. That framing matters, because it tells you exactly what can go wrong: if the history is wrong, the costs are wrong, or the window is cherry-picked, the rerun is wrong too — and it will still print a beautiful equity curve.

If you are still unsure what an Expert Advisor is or how one turns coded rules into live orders, settle that first — this page starts at the moment you already have a compiled file and need to prove it.

Everything the Strategy Tester needs from you beforehand fits on the checklist below, and two of its lines carry far more weight than the rest. The first is how your EA behaves inside a bar: whether it acts only at bar close, or manages stops, trails and pending orders while the bar is still forming, decides which testing model is honest for it — a decision the next few sections keep returning to. The second is the pass/fail bar you write down before you run, because numbers are persuasive after the fact; fix your minimum trade count, worst tolerable drawdown and minimum profit factor now, and let the report argue with your bar instead of the other way around. The remaining lines are mechanical, with one caveat worth knowing in advance: MT5 downloads history from the broker's server on demand while MT4 uses whatever already sits in its history centre, and thin history is the most common cause of a nonsense first result.

How to Backtest an EA in MT4 (Step by Step)

MT4 puts the whole tester on one strip at the bottom of the terminal. It is fast to configure and easy to misconfigure, because three of its most consequential settings sit next to each other with harmless-looking defaults.

Opening the Strategy Tester and Selecting Your EA

Press Ctrl+R, or use View → Strategy Tester. The panel docks along the bottom of the terminal.

Set the mode selector to Expert Advisor (the same panel also tests custom indicators), then pick your EA from the dropdown. If it is not listed, the file is either not compiled or not in MQL4/Experts — compiling it in MetaEditor and restarting the terminal populates the list.

Click Expert properties and open the Inputs tab. This is where the EA's own parameters live: lot size, risk percentage, stop distance, session filters, magic number. For a single test pass only the Value column matters; the Start, Step and Stop columns exist for optimisation runs. Load the values you intend to trade before anything else, because every number in the report is a consequence of them.

Setting Symbol, Period, and Date Range

Symbol lists the instruments your broker's feed exposes. Watch for suffixes — EURUSD, EURUSD.m and EURUSDmicro can be three separate symbols with three separate histories on the same server, and only one of them is the one you trade.

Period must be the timeframe the strategy was designed for. Running an H1 strategy on M15 does not "test it more thoroughly"; it tests a different strategy.

Tick Use date and set From and To deliberately. Two rules make the difference between a window and a cherry-pick:

1. Cover more than one market regime. A range that contains only a clean trend will bless every trend-following EA ever written. Include a choppy stretch and a volatile stretch.
2. Hold something back. Stop the test period short of your most recent history and leave that slice untouched. It becomes your out-of-sample test later — the single most informative thing you can do with a backtest.

The Visual mode checkbox replays the test on a chart with a speed slider and a Skip to button. Turn it on for the first two minutes to confirm the EA is opening the kind of trades you expect, then turn it off — visual mode is dramatically slower and you do not need to watch six years of history tick past.

Configuring Deposit and Leverage

Back in Expert properties, open the Testing tab. Two settings here shape the entire report:

• Initial deposit and currency. Set this to the account you would actually fund. Position sizing, margin availability and every percentage in the report are all relative to it. A 0.1-lot EA tested on a hundred-thousand-dollar deposit produces a drawdown figure so small it means nothing.
• Positions: Long & Short / Only Long / Only Short. Leave it on Long & Short unless the strategy is genuinely one-directional. Restricting it is also a useful diagnostic: if the whole edge disappears when you disable longs, the EA was riding a trend, not trading a rule.

MT4 has no leverage field. The tester inherits the leverage of the account the terminal is logged into, which is a trap worth naming: a demo account at 1:500 will happily open positions your live 1:30 account would reject on margin. Log into an account with the leverage you will really trade before you test.

The spread box on the main strip — and everything else that decides whether your costs are realistic — is deliberately not covered here. Those settings get their own section below, because they are the ones that separate a credible test from an expensive one.

Running the Test and Reading the Output Tabs

Two decisions still stand between you and a credible run, and each has its own section below: the Model dropdown on the main strip, which is the most consequential control in the whole panel, and the Spread box beside it. Set both deliberately instead of accepting what the tester offers, then press Start.

The progress bar fills from left to right, and Stop ends a run you can already tell is wrong. When the pass finishes, the tabs along the strip carry the output: Results lists every simulated trade in order, Graph draws the balance and equity curves, Report is the summary you will actually judge the EA on, and Journal records what went wrong during the run — worth a glance, because an EA that failed to open half of its trades still produces a tidy-looking report.

How to Backtest an EA in MT5 (Step by Step)

MT5's tester is a different machine, not a reskin. It is multi-threaded, it can test many symbols in one pass, it can pull real tick history from the broker, and it exposes execution settings MT4 simply does not have. That is more power and more places to get it wrong.

Opening the Strategy Tester and Choosing Single-Symbol vs Multi-Currency

Ctrl+R or View → Strategy Tester docks the panel at the bottom, organised into tabs: Settings, Inputs, Agents, Journal — with Backtest, Graph and results tabs appearing once a run finishes.

On the Settings tab, pick your EA in the Expert field (compiled .ex5 files from MQL5/Experts), then make two choices before anything else:

• Type of run. Keep it on a single test pass for now. The optimisation modes exist to search a parameter space, and searching a parameter space before you know the EA works unoptimised is the fastest known route to a curve-fitted result.
• Single-symbol or multi-currency. An MT5 test can reach every symbol in Market Watch, so a portfolio EA that trades eight pairs is tested in one pass — a genuine capability MT4 has no answer to. The flip side: if your EA trades one symbol, trim Market Watch first, or the tester will download history for instruments it never touches.

The Forward field deserves a moment. Set to 1/2, 1/3 or 1/4, MT5 automatically splits your date range and reserves the most recent portion as an untouched forward segment, reporting it separately. That is out-of-sample validation built into the platform, and it costs you one dropdown.

Setting Symbol, Period, and Date Range

Symbol and Period work as they do on MT4, with the same suffix caution and the same rule about testing the timeframe the EA was designed for.

Date offers presets — last month, last year, entire history — plus a Custom period with explicit From and To fields. Use the custom range: presets tempt you into whatever window flatters the strategy.

Be ready for the first real-tick run over a long range to take a while and to pull down a lot of data. MT5 requests tick history from the broker's server the first time you ask for it, then caches it. That wait is the price of the most honest simulation either platform can produce.

Configuring Deposit, Leverage, and Testing Parameters

The lower half of the Settings tab is where MT5 gives you what MT4 withholds:

• Deposit and currency — same reasoning as MT4: fund the tester like you would fund the account.
• Leverage — an explicit dropdown. Set it to your real account's leverage. It decides margin, and margin decides when the tester stops being allowed to open the next trade.
• Modelling — the tick-generation mode. It lives here, but the decision belongs to the next section; do not accept the default without reading it.
• Delays — execution realism, covered with the other cost settings below.
• Profit in pips — strips currency conversion out of the result. Useful for checking whether the rules fire correctly, misleading when you care about money. Leave it off for a verdict run.
• Visual mode — same purpose as MT4's, same advice: a short sanity check, then off.

Finally, the Inputs tab holds the EA's own parameters with Value / Start / Step / Stop columns. For a single pass, fill in Value and ignore the rest.

Running the Test and Reading the Output Tabs

Two settings are still outstanding, and both are decisions rather than defaults: Modelling and Delays, each covered in its own section below. Choose them, then press Start. MT5 hands the pass to a local testing agent rather than running it inside the terminal window, so the terminal stays usable while the test works through the range.

When the run finishes the tabs change: Backtest holds the report metrics, Graph draws the balance and equity curves, Results lists every deal in order, and Journal records what the tester and the EA reported while running — the first place to look when the trade count comes back surprisingly low. If you set a Forward split, MT5 reports that segment separately; read it next to the main one, because a result that holds on the tested stretch and falls apart on the reserved stretch has told you the most useful thing the tester can say.

Which Testing Model Should You Use?

This is the setting that decides whether your result is evidence or fiction, and it is the one most guides skip in a sentence.

The testing model controls how many price points your EA gets to see between the open and close of each bar. An EA that evaluates its rules once per closed bar barely notices the difference. An EA with an intrabar stop, a trailing stop, a pending order, or a scalping entry is a completely different program under each model — because the model determines whether the price ever "visited" the level that would have triggered it.

Three practical rules follow from that table:

1. If the EA does anything intrabar, "Open prices only" is not a fast test — it is a different strategy. Stops and take-profits that would have been hit mid-bar are evaluated against a single price. Results from this mode routinely look better than reality by a wide margin.
2. If the EA genuinely acts once per closed bar, "Open prices only" is legitimate and saves hours during early screening. Confirm the claim before you rely on it; "it only trades at bar close" is often true of entries and false of exits.
3. Every verdict run should be every-tick. On MT5 that means real ticks. On MT4 it means the every-tick model, and — if the result is going to decide real money — imported tick data rather than the platform's own M1 history.

MT4 reports its own honesty score in the report header as Modelling quality. Using the platform's built-in M1 history, the every-tick model tops out at 90%; only externally imported tick data pushes it higher, and the faster models report far lower. Check the accompanying mismatched chart errors count in the same header: a large number there means your M1 history has holes and the tester filled them with bars that never existed.

Matching the Backtest to Real Trading Conditions

Every cost the tester does not model is a cost your live account pays anyway. This is where most "profitable" EAs quietly die.

Spread. MT4's spread box defaults to Current — the spread at the exact moment you pressed Start, frozen across the entire history. Press Start on a Sunday evening and you have just tested years of trading at a weekend spread. Enter a fixed value in points instead, and choose a pessimistic one: not the broker's advertised typical spread, but something closer to what you meet around news and at the session roll. On MT5 with real ticks the historical spread arrives with the tick data, which is one more reason to prefer that mode.

Commission. MT4's tester has no concept of commission at all. None. An EA scalping a few pips per trade can print a rising equity curve in MT4 and lose money from day one on a raw-spread account. You have two options: set the tester's spread high enough to swallow your round-turn commission, or subtract commission per trade from the report by hand. MT5 handles this properly through the symbol's contract settings.

Slippage and execution delay. MT4 fills instantly at the modelled price — no delay, no requote, no partial fill. MT5's Delays setting is the closest either platform comes to modelling reality: No delay, Random delay, or a fixed value in milliseconds. Run at least one pass with a delay applied. If a strategy's edge disappears when orders arrive a fraction of a second later, that edge was never yours to collect.

Swap / rollover fee. Applied from the symbol's properties, and easy to forget until an EA that holds positions for days turns out to have been paying for the privilege the whole time.

Now do the arithmetic that kills bad EAs fastest. Suppose your report shows 600 trades and an expected payoff equivalent to two pips per trade, and your realistic all-in cost is the two and a half pips above. The strategy does not have a small edge — it has a negative one, and the tester simply did not charge you for it. Any EA whose average win is smaller than its true transaction cost is a cost generator with extra steps.

How to Read Your Backtest Report

Read the report in three passes: header first (is this test even valid?), then the risk lines, then the profit lines. Reading it top to bottom is how people fall in love with a net profit figure produced by a broken test.

Header — validity. Modelling quality, bars in test, ticks modelled, mismatched chart errors. If quality is low for an intrabar EA, or the mismatch count is large, stop. Nothing below the header means anything yet.

Total net profit. The line everyone reads first and the one that decides least. It answers how much, never how reliably, and it is the figure most inflated by a test that charged no costs. Read it last.

Total trades. The statistical foundation. Under roughly thirty trades you are reading noise with decimal places; a few hundred is where the numbers start describing a process rather than a coincidence. Everything else on the report inherits this line's credibility — which is why sample size is the first thing to check after validity.

Profit factor — gross profit divided by gross loss. Above 1.0 is profitable on paper. Your bar is not "above 1.0" but "comfortably above 1.0 after subtracting the costs the tester ignored, and still above 1.0 when you shift the date window." A profit factor several times higher than that, on a thin trade count, is a warning rather than a trophy.

Expected payoff — average money made per trade, which is expectancy expressed in account currency. This is the line that survives cost subtraction or doesn't. Deduct your real per-trade commission and slippage from it; if what remains is close to zero, you have no strategy.

Absolute, maximal and relative drawdown. Relative drawdown — the deepest fall as a percentage of the balance peak — is the one that maps to your tolerance. Set your bar honestly: assume the live version will be deeper than the tested worst case, because it usually is, and ask whether you would still be running the EA at that depth. It helps to see what a given drawdown demands in return before you accept it — our drawdown recovery calculator turns the percentage into the gain required to get back to flat.

Profit trades, and the long/short split. A historical win rate on its own tells you nothing without the average win against the average loss. The directional split is quietly informative too: if essentially all the profit comes from long trades over a decade-long uptrend, you tested a market direction, not a rule.

Largest profit trade versus average profit trade. Mentally delete the largest win. Is the EA still profitable? If one trade carries a large share of the gross profit, the result is a lottery ticket that has already been cashed.

Maximum consecutive losses — in both count and money. Multiply the count by your risk per trade to see the losing streak the strategy expects you to sit through, then ask whether you would actually sit through it. Most EAs are switched off during a normal losing streak, not a catastrophic one.

Recovery factor — net profit divided by maximum drawdown, i.e. how much the strategy earns per unit of pain. And on MT5, the Sharpe ratio tells you how smooth the ride was rather than how far it travelled.

How Do You Spot an Overfit EA From the Report?

Overfitting — curve fitting — is a strategy tuned so precisely to one stretch of history that it has memorised it instead of learning anything about markets. It is the reason the field is full of EAs with spectacular backtests and dismal live records, and the report will tell you about it if you know which lines to distrust.

Seven red flags, in the order they usually appear:

1. Too few trades. A dazzling result over forty trades is a coincidence with good marketing. This is the flag that invalidates all the others.
2. An equity curve with almost no dips. Real edges lose regularly and in clusters. A near-straight diagonal line is not a sign of quality; it is a sign that losses were engineered out of the sample.
3. Profit concentrated in a few trades or one period. Check the largest-win line and the graph shape. If the curve is flat except for one vertical step, you tested one event.
4. Results that evaporate when the window moves. Shift the start and end dates by a month in each direction and re-run. A robust strategy degrades; an overfit one collapses. This costs two minutes and is the cheapest overfitting test in existence.
5. A long list of finely-tuned inputs. A 17-period average, a 2.7 ATR multiplier and a 43-pip filter did not come from a market principle. They came from a search that found the exact combination that fit this data.
6. It only works on one symbol and one timeframe. Test a neighbouring pair and the adjacent timeframe. You are not looking for equal performance — you are looking for the edge to still be visible.
7. A near-perfect win rate with small average wins. This is the fingerprint of martingale or grid recovery logic. Check the largest loss trade and the money value of the maximum consecutive losses: the entire risk of these systems hides in a single column of the report.

The counter-move is simple and non-negotiable. Re-run the identical configuration on the date range you deliberately held back — untouched history the parameters were never chosen against. On MT5 the Forward field does this for you. If the reserved slice tells the same story as the tested slice, you have something. If it doesn't, you have a memorised chart.

Two Reports, One Passes: The Checks Applied

Criteria only become a skill once you have watched them applied. Below are two invented reports for the same kind of strategy. One of them is the one every marketplace screenshot looks like. The other is the one you would actually run.

The uncomfortable lesson is that the report you want to see and the report you should want to see look nothing alike. A tested edge that is small, noisy and consistent is something worth forward testing. A tested edge that is enormous, smooth and rare is almost always a settings artefact — and the settings are yours to fix.

MT4 or MT5 for Your Verdict Run?

Report B passed on conditions MT5 hands you and MT4 asks you to assemble by hand: real ticks that carry the historical spread with them, commission charged through the symbol's contract settings, an execution delay you can actually apply, an explicit leverage field, and a reserved forward slice. If your EA has both builds, that settles the question — screen on whichever is convenient, but run the test that decides real money on MT5.

When only an .ex4 exists, the gap is closable; it just is not closed by default. Import third-party tick data instead of trusting the built-in M1 history, so modelling quality is no longer capped by the platform's own ceiling. Log the terminal into an account carrying your live leverage before you press Start, because the tester silently borrows it. Widen the fixed spread until it swallows your round-turn commission, and treat the missing execution delay as an unpriced risk rather than a solved one — which in practice means demanding a wider margin above your profit-factor bar on MT4 than you would accept on MT5. And because MT4 tests one symbol per pass, a portfolio EA has to be run pair by pair: add those drawdowns up by hand rather than assuming the worst ones politely land in different months.

One more difference changes results rather than presentation: MT5 accounts can be netting or hedging, and MT4 accounts are always hedging. On a netting account, a buy while short reduces the existing position instead of opening a second one — so an EA that assumes independent positions can behave differently under test than you expect. Check which account type the test is running against, and see hedging vs netting if the distinction is new to you.

What to Do After a Good Backtest

A clean backtest earns your EA one thing: permission to be tested in real time. It does not earn it your money.

The rung that matters most is the second one. A forward test puts the same EA in front of a live feed on a demo account, where every assumption the tester made politely on your behalf gets contested by reality: the spread widens at the exact moment your entry triggers, an order is rejected, a weekend gap jumps your stop. Run it long enough to matter — if the backtest averaged three trades a week, a fortnight is not a forward test, it is a mood.

Two adjacent topics belong here as signposts, not detours. The Optimization tab is the natural next step once a single unoptimised pass looks credible — and it is also the fastest way to overfit a strategy, which is exactly why it comes after this stage rather than before. And VPS hosting for an EA becomes a real question only once something has earned continuous uptime — how to choose one is a decision of its own, and a strategy that has not passed a forward test does not need 24/7 anything.

The honest summary of everything above: the Strategy Tester never tells you whether an EA is good. It tells you what your rules did to a specific slice of history under specific assumptions, and the whole skill is in choosing assumptions that do not flatter you. Set the bar before you run, charge the test full price, hold history back, and let the report be the first thing that disagrees with you — long before your account does.

FAQ

How many trades does a backtest need before the result means anything?

There is no magic threshold, but the practical shape is clear: below roughly thirty trades you are measuring luck, between thirty and a hundred you have a hint, and from a few hundred upward the metrics start to describe the strategy rather than the sample. Trade count also has to be earned honestly — extending the date range to inflate it is fine, loosening the entry filter until more trades appear is just a different way of fitting the curve.

Why do my backtest and my live results look nothing alike?

Almost always one of four things: the testing model let intrabar rules fire at prices the market never really offered; commission and slippage were never charged; the spread used in the test was narrower than the one you actually trade; or the strategy was tuned to the exact window it was tested on. Work through them in that order. If the gap survives all four checks, the remaining suspect is execution — order rejections, requotes and latency that no tester fully reproduces.

Can I backtest an MT4 EA on MT5, or the reverse?

No. An .ex4 runs only on MT4 and an .ex5 only on MT5, and neither tester loads the other's format. Many developers publish both builds of the same strategy, but they are separate compilations — and because the two testers model costs and ticks differently, the same strategy will not produce identical reports on both. If both builds exist, run your verdict test on MT5.

Does 90% modelling quality mean my backtest is 90% accurate?

No. Modelling quality describes how completely the platform was able to reconstruct price movement inside each bar from the history it had — it says nothing about whether your costs, spread, execution assumptions or date range were sensible. A test can report the highest possible modelling quality and still be worthless because it charged no commission or covered one market regime. Treat it as a validity check on the data, not a confidence score on the result.

How far back should the test period go?

Far enough to contain market conditions that genuinely differ from each other, and long enough for the strategy's natural trade frequency to produce a meaningful sample. For an intraday EA that may be a couple of years; for a swing strategy taking a handful of trades a month, considerably more. Then hold back the most recent stretch entirely, so you keep a slice of history the parameters were never chosen against.

Is running the Optimization tab the same as backtesting?

No, and conflating them is how good strategies get ruined. A backtest asks how one fixed configuration performed. Optimisation runs hundreds of configurations and hands you the one that performed best on that exact data — which is a description of the past, not a prediction. Establish that the EA works with sensible unoptimised settings first; only then is a parameter search a refinement rather than a curve-fitting exercise.

Do I still need to backtest an EA that comes with a published track record?

Yes, for a reason that has nothing to do with trusting the seller: their result was produced on their broker's spread, their symbol, their leverage and their execution. Yours will not be. Re-running the test on your own feed, with your own costs and your own account settings, tells you what the strategy does in the conditions you will actually trade — which is the only version of the answer that helps you.

Sources & Further Reading

Want to go deeper? These independent, authoritative sources shaped this guide — each one is worth reading in full:

• MetaTrader 5 Help — Real and Generated Ticks: how each modelling mode builds price movement
• MetaTrader 5 Help — Strategy Testing: every setting in the MT5 tester explained
• MetaTrader 4 Help — Strategy Tester Report: what each core report metric means
• MQL5 Articles — Testing trading strategies on real ticks
• Wikipedia — Overfitting: why a model that fits its sample too well fails on new data