Why a gold EA backtest differs from live trading

A backtest is a model of your broker, and a live account departs from it on spread, commission, fills, tick data and connection, each in a known direction that you can name and, in most cases, measure.

• A backtest is a model of your broker, not your broker. Every gap between the two can be named, and most can be measured.
• The two we publish on every figure: the tester charged no commission on XAUUSD, and its fills are modelled rather than executed.
• On our own Dawnbreak run, adding back roughly $7 per lot round-turn takes about $877 off the Strategy Tester net. Do that arithmetic on any EA you consider.

Search for why an EA backtest differs from live results and you will find forum threads full of the same surprise, and vendor pages that answer it with a rule of thumb nobody can source. There is a better answer: the MetaTrader 5 Strategy Tester replays the past under a set of stated assumptions, and a live account breaks each one in a known direction. Name them and the gap stops being a mystery.

A backtest is a model of your broker

The tester does not trade. It reads stored prices, decides where your orders would have filled, and books a result. Five of its assumptions matter on gold: the spread it applies, the costs it charges, how it fills an order, the quality of the ticks it replays, and the fact that it never loses its connection. A live account differs on all five, and the tester’s version of each is the tidier one.

Spread: recorded, or smoothed

XAUUSD has no exchange-set spread; your broker quotes it, and it moves. Whether the tester sees that movement depends on the modelling mode. MetaQuotes’ documentation on tick generation puts it plainly: “When testing on real ticks, a spread may change within a minute bar, whereas when generating ticks within a minute, a spread fixed in the appropriate bar is used” (MetaTrader 5 Help).

So a run on generated ticks gives a whole minute one spread, and the minute in which a release hits the market is exactly the minute where that flattening flatters. Gold EAs and NFP, CPI and FOMC shows how to find those minutes in your own run.

Commission: the cost the tester did not charge

This is the largest gap we can put a number on, and we publish it on every figure. On the IC Markets demo feed the desk tested, the MetaTrader 5 Strategy Tester charged $0 commission on XAUUSD. Real trading on that account costs roughly $7 per lot round-turn, and the net on every one of our strategy pages does not deduct it.

What it costs depends on how much a strategy trades, not on how well it did. Two of our own published runs, both at a $25,000 deposit over 1 January to 7 September 2026, from our own Strategy Tester runs:

• Dawnbreak closed 556 trades and 125.22 lots. The Strategy Tester net was +$12,518.94; at $7 a lot the commission comes to about $876.54, leaving about $11,642.40 — roughly 7 % of the net gone.
• Reclaim closed 65 trades and 8.81 lots. Net +$17,222.66; commission about $61.67, well under 1 %.

The same fee, two very different bites. A strategy that trades often pays for every one of those trades in live conditions, and the tester leaves the bill out. The Dawnbreak strategy page states the departure beside its figures; the arithmetic above is simply that departure worked through, and gold EA costs per lot shows how to make the tester charge it.

Fills: the tester decides the price

In the tester, where an order fills is a setting. MetaQuotes describes the No Delay option as one where “all orders are executed at requested prices without requotes. The mode is used to check how an EA would perform in ‘ideal’ conditions” (MetaTrader 5 Help). There is also a Random Delay mode and a fixed one, which push a run towards the waiting a live order does.

A live account works differently. Under Market Execution, “a broker makes a decision about the order execution price without any additional discussion with the trader”, and “Sending an order in such a mode means advance consent to its execution at this price” (MetaTrader 5 Help). Under Instant Execution the server can refuse and offer a new price instead: a requote. An Expert Advisor can cap how far it will accept being moved through the request’s deviation field, “The maximal price deviation, specified in points” (MQL5 reference), but a cap trades slippage for missed entries; it does not remove the gap.

That is why our second published departure reads: fills are modelled, and a live account is filled by a broker at a price that can be worse, and around news much worse.

Tick data: how much of the window is real

The tester can only replay the ticks it has. Our own campaign came back at History Quality 63 %, because IC Markets served real ticks only from 21 April 2026, so January to April were synthesised from M1 bars. Every figure we publish over that window carries the caveat, and every conclusion about those four months is an approximation. How to backtest a gold EA in MT5 explains how to read the number.

A second point follows from it: your broker is not our broker. Different feeds record different spreads, and a symbol with a different suffix can be a different contract. A run on one feed does not transfer to another.

The machine it runs on

The tester never disconnects, never restarts for an update and never sits in a queue behind a slow network. A live Expert Advisor can manage a position only while its terminal is running and connected, which is what a VPS for a gold EA on MT5 is for. Latency narrows one gap among five; it changes neither the spread nor the commission.

How to close the gap before you trade

• Test on Every tick based on real ticks, on your own broker’s symbol, and read History Quality first.
• Run once on No Delay and once on Random Delay. A strategy whose result depends on the difference is telling you something.
• Take the volume from the report’s deals, multiply by your broker’s per-lot commission, and subtract it from the net yourself.
• Read the drawdown on equity, not balance, and at the deposit you will trade.

The free Backtest edition of each Product exists for this: it starts only inside the Strategy Tester, so every assumption above is yours to set. Pick one from the strategies.

Before you act on any of this

Every figure above comes from the MetaTrader 5 Strategy Tester over 1 January to 7 September 2026 on IC Markets demo data, never from live trading, and no result is promised anywhere on this site. The commission arithmetic is our own estimate at roughly $7 per lot, not a broker statement. Read the risk disclosure before you attach anything to a funded account.