Spent the higher a part of final yr constructing Nexus 9 — a 9-pair imply reversion EA — and found alongside the best way that MetaTrader 5’s Technique Tester silently produces deceptive outcomes for multi-symbol methods until you modify a handful of non-default settings.
When you have ever:
- Run a multi-symbol backtest that appeared too clear, then watched it disintegrate reside
- Obtained an fairness curve that appeared flat or bizarre and couldn’t work out why
- Questioned why your EA “doesn’t commerce” sure pairs in tester
… this publish is the guidelines I want I would had after I began.
The core drawback
The MetaTrader 5 Technique Tester was initially designed round single-symbol testing. Multi-symbol assist exists, however it’s gated behind a number of non-default settings. In case your EA accesses pairs apart from the chart image, by default:
- Tick historical past for non-chart pairs is not going to load
- SymbolInfoDouble() returns zero or stale costs for non-chart symbols
- Trades on these symbols silently fail to register
Your fairness curve will look easy, your stats will look acceptable, and the info feeding your EA will probably be improper. The tester provides no warning.
Required setting #1: Market Watch → Present All
Earlier than opening Technique Tester, right-click within the Market Watch panel → “Present All”.
This tells MT5 to make all out there symbols candidates for tick knowledge loading. With out this, solely at present seen symbols get knowledge — and any pair your EA references that’s hidden returns clean ticks.
Required setting #2: Pre-load tick historical past per image
Even after “Present All”, MT5 doesn’t robotically obtain historical past. For every image your EA trades:
- Proper-click → “Image Specification” → affirm enabled
- Open a chart for that image (any timeframe — opening forces obtain)
- Anticipate the progress indicator on the bottom-right to finish
For 9 pairs, that’s 9 guide chart-opens. Tedious. However skip it and you might be testing on holes within the knowledge.
Required setting #3: “Each tick based mostly on actual ticks”
Technique Tester defaults to “Each tick” — the generated tick mannequin. Nice for single-symbol on liquid pairs. Ineffective for multi-symbol as a result of the generator assumes a single chart context. Cross-pair correlations and shared timing home windows don’t reproduce.
Swap to “Each tick based mostly on actual ticks”. It makes use of precise dealer tick knowledge throughout all symbols in right chronological order. About 10× slower. The one mode that provides you accuracy.
Required setting #4: Confirm the dealer’s server time zone
Multi-symbol EAs usually base entries on session instances (London / NY / Tokyo). Brokers report bar timestamps within the dealer’s server time zone — Exness makes use of GMT+3, IC Markets makes use of GMT+2 with DST, and so forth.
In case your EA hard-codes hours like if (hour == 21) , these imply various things on totally different brokers. Earlier than testing:
- Open M1 chart in your dealer
- Hover over a candle to see its timestamp
- Evaluate to precise UTC
Doc the offset — both auto-detect within the EA, or expose it as a configurable enter.
Pitfall #1: the Unfold parameter
Defaults to “Present” — the unfold proven proper now in your terminal. Nice for brief checks. Wildly unrealistic for multi-year backtests the place unfold different dramatically.
Both set unfold to a sensible worst-case fixed, or use the real-ticks mode (which preserves historic unfold variation per tick). A check that ignores unfold variability will systematically overestimate your edge — and for tight-range methods, can flip a winner right into a loser.
Pitfall #2: optimization runtime explodes
Multi-symbol × parameter mixtures × years of knowledge = days of compute, even on quick machines. Pragmatic strategy:
- Optimize on EURUSD or one other consultant pair first
- Validate the chosen parameters on all pairs in common backtest mode
- Ahead-test on demo earlier than going reside
Pitfall #3: VPS is the improper place to backtest
Counter-intuitive, however buying and selling VPS plans (1-2 GB RAM, CPU profiles tuned for low-latency execution) are horrible for batch tick processing. Multi-symbol backtests want 8-16 GB RAM. On a 1 GB VPS, the check swaps to disk and runs 10-100× slower.
Backtest regionally. Use the VPS for reside deployment.
Pre-flight guidelines
Earlier than urgent Begin in Technique Tester:
- Market Watch → Present All
- Tick historical past downloaded for each pair (chart opened)
- Tick mannequin = “Each tick based mostly on actual ticks”
- Unfold set realistically (not “Present” for lengthy historic checks)
- At the least 8 GB RAM on the check machine
- Dealer server time zone verified
- Operating regionally, not on a low-end VPS
If any are lacking, the outcome is just not reliable. Don’t make buying and selling choices on unhealthy knowledge.
What I constructed utilizing this system
Nexus 9: 9 FX pairs (EURUSD, EURGBP, EURJPY, GBPJPY, USDCHF, EURAUD, GBPAUD, GBPCAD, EURCAD) + XAUUSD as a tenth instrument. Imply reversion at session boundaries. Mounted SL/TP on each commerce. No martingale, no grid, no averaging.
11-year backtest (01 Jan 2015 → 31 Mar 2026), $1,000 beginning steadiness → $18,508 ending fairness at 37.39% most drawdown. Each check run with the settings above.
Full fairness curve, parameters, and version comparability on the mission web page:
MQL5 product pages:
A separate Neighborhood Version (account-bound, free beneath our Exness IB partnership, XAUUSD at all times enabled) is offered by means of the mission website.
If you’re constructing your personal multi-currency system, hopefully this guidelines saves you some ache. The largest lie in retail systematic buying and selling is a ravishing fairness curve from a misconfigured Technique Tester. Backtest accurately.
