Articles · Mike Harty · Published 2026-10-24 · Updated 2026-10-24
How to Backtest an MT5 EA Properly: Ticks, Costs, Set Files
Short answer: Test on your own broker's MT5 server using "Every tick based on real ticks", over at least five years if the data exists. Load the EA's real .set file, use the lot size and deposit you'll trade, and check History Quality, commission and swap in the report. Then judge it on equity drawdown, not on profit.
Step 1: Use your broker's data
Open the Strategy Tester (Ctrl+R) in the terminal logged in to the broker and account type you'll trade. Spreads, commission, swap, contract size and trading hours all come from that server, so a test on another broker's data is a test of different conditions. Ideally, use the account type (raw or standard) you'll run.
Step 2: Choose the modelling mode
| Mode | What it does | Use it for |
|---|---|---|
| Every tick based on real ticks | Replays the broker's recorded ticks and spreads | Final tests, and any grid, scalping or recovery EA |
| Every tick | Generates ticks from M1 bars | Only when real ticks aren't available |
| 1 minute OHLC / Open prices only | Four or one prices per bar | Quick optimisation passes, never final numbers |
After the run, read History Quality in the report. It shows how much of the test used real ticks. Our tests have ranged from 100% to 19%. At 19%, most ticks were generated, so we treat that result as lower confidence and say so next to it.
Step 3: Set the period, deposit and leverage
- Period: cover several market regimes, for example 2020 to today. Short tests hide the rare bad months that size your capital.
- Deposit: the amount you'll actually fund. Margin stops depend on it.
- Leverage: your account's leverage, because that sets when margin calls would have happened.
Step 4: Load the real set file
On the Inputs tab, right-click and choose Load to open the vendor's or your own .set file. Check the following before you run: - the lot or risk setting (fixed lot versus auto lot) - the maximum number of orders or grid levels - any stop-loss, equity-stop or basket-stop inputs - slippage, spread and time filters
The report prints the inputs it used. Check that they match the set file. If the EA uses auto lots, confirm that the deals show the lot sizes you expected.
Step 5: Test at the lot size you'll trade
A 0.01-lot baseline is useful for comparison, but it doesn't always scale. Multi-order and recovery EAs can split or grow positions differently at larger lots. In our tests of a popular MQL5 recovery EA, scaling the 0.01 run ×17 predicted a $3,322 worst drawdown. The real 0.85-lot run hit $8,399.39 of equity drawdown (GBPUSD H1, 2020–Oct 2026, 100% history quality, backtest not live). Run the real lot as well.
Step 6: Read costs in the report
Note the commission and swap totals. On one 0.85-lot run of ours, these came to $11,028 and $2,809. For a high-frequency EA, costs can be the difference between profit and loss. If you test on a raw-spread account, commission must appear. If it shows $0, the tester isn't charging it.
Step 7: Check the report, in this order
- History Quality and the period actually tested.
- Equity drawdown, maximal ($ and %). This is the one that sizes capital. See reading MT5 backtest drawdown.
- Did the test end flat? Open positions at the end aren't in net profit. One of our tests ended on a −$14.8k open basket, so its headline profit overstated the result.
- Worst month and longest losing streak (from the deals list).
- Profit factor and net profit last.
What a backtest can't tell you
- Execution: the tester doesn't reproduce real delays, requotes or slippage. Add them afterwards. In our Monte Carlo and slippage stress of four EAs (5,000 resamples each, 24 Oct 2026), adding +2 pips per basket cut one grid EA's median net from $36,410 to $6,567, while its P95 drawdown went from $29,201 to $36,424. That EA's backtest ended on an open −$14.8k basket. Its 86% blow % means its worst single event is bigger than the $4,000 floor. It is not a forecast of how often the account will blow. We haven't modelled execution delay.
- Live behaviour: only a live or demo account shows whether the EA trades like its backtest.
- Curve-fitting: a heavily optimised .set can fit the past perfectly. Prefer settings that hold up across nearby values and periods.
For our full process, see how we backtest EAs. To turn a report into a capital figure, see how much capital you need to run an EA.
Start free: VRD Insight
Once the EA is live, VRD Insight (free, on the MQL5 Market) tracks each EA by magic number so you can compare it with the backtest. Insight's drawdown is calculated from closed trades, so watch floating equity on grid or recovery EAs as well. VRD Quant ($69) turns a backtest into a capital plan, and VRD Complete ($149) bundles Insight, Guardian and Quant.
Risk disclaimer: Backtests are historical and hypothetical, not live results, and do not predict future performance. Nothing here is investment advice. Trading leveraged products carries a high risk of loss.
Frequently asked questions
Which modelling mode should I use to backtest an EA in MT5?
'Every tick based on real ticks'. It replays the broker's recorded ticks and spreads. Check the History Quality figure in the report. Anything well below 100% means many ticks were generated, so treat the result with less confidence.
Does the MT5 Strategy Tester include spread and commission?
With real ticks, it uses the recorded spread of each tick. Commission and swap come from the symbol and account settings of the broker server you test on, so test on the broker you'll trade with and check the commission total in the report.
How do I load an EA's settings into the Strategy Tester?
On the Inputs tab, right-click and choose Load to open the .set file. That way you test the settings you'll actually run, not the defaults. Save the .set alongside the report.
Can I backtest at 0.01 lots and scale the result up?
Only for simple single-position EAs. Recovery and multi-order EAs behave differently at larger lots. In one of our tests, the real-lot drawdown was about 2.5 times what scaling the 0.01 result predicted.
Is a good backtest proof that an EA works?
No. A backtest is historical and hypothetical. It doesn't include real execution delays or slippage, and it can be curve-fitted. Use it to size risk, and then check that live behaviour matches.
One EA blowup costs more than the whole suite.
VRD Complete contains single-EA blowups and frees capital you have parked as idle buffers. $149 at launch pricing.