Expert advisors (EAs): what they are, why most fail, and how to evaluate one honestly

What is an expert advisor?

An expert advisor (EA) is a program that runs inside the MetaTrader platform and trades the account automatically. Written in MQL4 (for MT4) or MQL5 (for MT5), an EA watches incoming prices, applies its coded rules, and places, manages and closes orders without manual input, including position sizing, stop losses and profit targets.

The appeal is real: perfect rule discipline, no emotion, no screen time, and coverage of sessions an Australian trader would otherwise sleep through. The London-New York overlap, the most liquid window of the forex day, runs through the middle of the night AEST, and an EA does not care.

What an EA is not: a source of edge. It is an execution vehicle for a strategy, and it will execute a bad strategy with exactly the same tireless discipline as a good one. That distinction drives everything else on this page.

Disclosure: SatoshiMacro may earn a commission if you open a broker account through links on this page, at no extra cost to you. Commissions never influence our testing-based rankings. See our full affiliate disclosure.

How EAs actually work on MT4 and MT5

The mechanics, briefly, because they matter for the evaluation section:

• The EA attaches to a chart in the MetaTrader terminal and receives every tick for that instrument. Its code runs on each tick or bar, checks its entry and exit conditions, and calls the platform's trade functions when conditions are met.
• It only runs while the terminal runs. Close the laptop and the EA stops managing its positions. This is why serious EA operation uses a VPS (a small always-on server, roughly USD 5 to 30 a month, free at some brokers past a volume threshold) so the terminal stays connected 24/5 with low latency to the broker.
• MQL4 and MQL5 are different languages. MT4 EAs do not run on MT5 without porting. The developer ecosystem has consolidated on MQL5, and MT5's strategy tester is decisively better: multi-currency backtests, real tick data, and cleaner forward-testing. New automated-trading setups in 2026 should be on MT5; the MT4 vs MT5 comparison covers the detail.
• The strategy tester is where EAs are born and where traders are misled. MetaTrader's built-in backtester runs an EA over historical data and produces the equity curves that populate every EA sales page. The tester is a legitimate tool; the way it gets used is the problem, covered next.

Why most purchased EAs fail live

The uncomfortable base rate: the overwhelming majority of commercially sold EAs lose money in live trading, and the mechanism is well understood.

• Overfitting is the killer. MetaTrader's optimiser can tune an EA's parameters across thousands of combinations until the historical equity curve looks perfect. That process does not discover an edge; it memorises the past. The more beautiful the backtest, the more likely it has been curve-fit, and the faster it degrades when market conditions shift from the fitted period.
• Backtests ignore real costs by default. Fixed-spread assumptions, no slippage, no swap, no spread widening at news. A high-frequency EA that backtests profitably at 0.5 pips of assumed cost can be structurally unprofitable at real all-in costs, which is why the spread mathematics matter doubly for automated strategies.
• Martingale and grid logic hide in bestsellers. Many top-selling EAs produce months of small steady wins by doubling into losing positions (martingale) or stacking grids. The equity curve looks serene until one trending move erases the account. A years-long win-rate claim with no large drawdown visible is a signature of this logic, not evidence against it.
• Survivorship does the marketing. Sellers launch many EAs; the ones whose live periods happened to go well get promoted, the rest quietly vanish. The testimonials are real, the inference is wrong.

None of this means automation is a scam. It means the burden of proof sits entirely on live evidence, which is the point of the checklist below.

The honest EA evaluation checklist

Apply all six before any purchase or live deployment. An EA failing two or more is a no.

1. Verified live results, not backtests. Demand a multi-year track record on a third-party verification service (Myfxbook or similar) with broker-linked, unmodifiable data. Screenshots and backtest curves are trivially fakeable and count for nothing.
2. Live period spans different regimes. A record that starts in 2024 has only seen one kind of market. Two-plus years spanning trending and ranging conditions is the minimum meaningful sample.
3. Drawdown honesty. Every real strategy has losing periods. A record showing 90-plus percent win rate with tiny drawdown is the martingale signature; look at the open-drawdown line on the verification page, not just closed-trade results.
4. Explained logic. You do not need the source code, but you need to know the strategy family (trend, mean reversion, breakout, news) and what conditions hurt it. "Secret algorithm" is a marketing phrase, not a strategy description.
5. No profit promises. Guaranteed returns, "passive income" framing, and monthly-percentage promises are the reddest flags in retail trading, and in Australia, unlicensed sellers promising returns sit outside every accountability structure.
6. Your own forward test. Whatever the record shows, run the EA on a demo account at your own broker for at least a month, then live at minimum size. Broker execution, spreads and slippage differ enough that the seller's results are not your results.

The same checklist, minus the code, applies to signal services and copy-trading providers, and the same risk-reward mathematics governs whether any win rate actually profits after costs.

Where legitimate EAs come from

Four sources, in rough order of trustworthiness:

• Your own build. Coding an EA in MQL5 (or assembling one with MT5's wizard) around a strategy you already trade manually is the highest-integrity route, because you know exactly what the rules are and why. The overfitting trap still applies to your own optimising.
• The MQL5 Market. The official marketplace built into MetaTrader lists thousands of EAs with user reviews and, importantly, rental options that let you test cheaply. Quality varies enormously; the checklist above is the filter, not the star rating.
• Freelance development. MQL5's freelance section and independent developers will code your specification, typically from about USD 100 for simple logic. You own the strategy; they write the implementation.
• Commercial EA vendors. The external websites with lifestyle marketing and countdown timers. A small minority are honest operations with verified records; the majority are the reason this page's failure section exists. Default to no.

Broker requirements: which AU brokers support EAs

Every major ASIC-regulated MetaTrader broker permits EAs, so the differentiators are execution quality, cost, and infrastructure:

• Pepperstone is the strongest all-round EA venue: raw Razor spreads, consistently fast execution (latency matters for EAs), MT4 and MT5 plus cTrader for its own automation ecosystem, and VPS support.
• Fusion Markets wins on cost at AUD 2.25 per side, which for high-frequency EAs is the largest controllable performance input.
• FP Markets offers competitive raw pricing with MT4/MT5 and VPS options.
• AvaTrade supports EAs on MT4/MT5 with spread-only pricing that suits lower-frequency automated strategies.

The best MT5 brokers comparison ranks the field in depth. One practical note: run any EA on the exact account type you will trade live, because an EA profitable on raw-plus-commission pricing can be unprofitable on a wider standard spread.

Frequently asked questions