Expert advisors: what to check before you buy one

Where a robot is allowed to run

Which brokers permit automated trading, what each says about it, and whether a VPS comes with the account. An expert advisor only trades while the terminal is running and connected, so this table decides more about a robot's results than the robot does.

Scroll the table sideways →

Read from each broker's own pages on 24 August 2026. Where a cell says a policy is not published, the broker may still permit automation — it simply does not say so anywhere we could read, which is worth asking about before you rely on it.

Why there is no top ten here

Because we would have to invent it. Ranking robots requires performance data, and the performance data that circulates is screenshots, backtests and equity curves supplied by the person selling the product. None of that survives contact with the question “on whose account, over what period, at what drawdown, with which broker's spreads”.

The CFTC puts the general case plainly in its own advisory: automated programs that provide buy or sell signals may help with trading discipline, but no technology can consistently predict the future. And the seller is paid whether or not the robot works, which is the same incentive structure our scam page describes for signal sellers.

So this page compares the two things that can be checked — what a real track record must show, and where a robot is allowed to run — and leaves the ranking to sites that are willing to publish one without evidence.

What a checkable track record looks like

MetaQuotes runs a monitoring service for exactly this, and the fields it publishes are the checklist: growth, profit per month, maximum drawdown, trading activity, share of profitable trades, initial deposit, trading time in weeks, trades per week, subscribers and subscription fee. A claim that cannot be expressed in those terms is a claim about nothing.

Three of those fields do most of the work. Maximum drawdown says how bad it got, which the equity curve in an advertisement never does. Trading time in weeks says whether the record covers a market that changed. Initial deposit tells you whether the percentages you are being shown came from a hundred dollars or a hundred thousand.

The last check is whose account it is. A monitored record on a live account at a named broker means something; the same numbers on a demo mean the fills were simulated, and simulated fills are exactly where a robot's edge tends to live.

Where expert advisors actually come from

MetaQuotes runs a marketplace for MT4 and MT5 products with ratings, reviews and rental pricing, and its filters tell you what buyers rely on: at least four stars, having reviews, offering rent. Renting before buying is the closest thing to a trial that exists in this market, and paid products can be downloaded free as a demo version to run in the strategy tester first.

The alternative is the private route — a seller on Telegram, Instagram or a forum, a file, and a bank transfer. Our scam catalogue lists that pattern for a reason: no ratings, no reviews, no refund, and a program that will hold your trading credentials.

Neither route makes a robot profitable. What the marketplace adds is that the claims are attached to an identity, a review history and a price you can compare.

The broker decides more about the result than the robot does

First, whether automation is permitted at all. FxPro states Expert Advisors are fully compatible with its MetaTrader platforms; Tickmill marks Classic and Raw as allowing all strategies while barring EAs on its $30 welcome account — the same broker, two rules, decided by which account you are on.

Second, whether it can stay running. An EA trades only while the terminal is open and connected, so a robot on a laptop stops when the laptop sleeps. RoboForex lists a free VPS among its services, Alpari and LiteFinance both offer one; where a broker does not, you are renting one yourself.

Third, the language. MQL4 runs on MetaTrader 4, MQL5 on MetaTrader 5 and cTrader Algo on cTrader, and none of them converts into another. If you already own a robot, that single fact eliminates most of the market before any other criterion applies.

And fourth, cost. A robot that trades often is a robot that pays spread and commission often: the difference between 0.2 pips and 1.6 pips per round turn is the difference between a strategy that clears its costs and one that does not, whatever the backtest said.

What testing proves, and what it does not

A backtest proves that the parameters fit the past, and the person who fitted them chose both. Run one yourself in the strategy tester rather than reading the seller's, and run it on the broker and spreads you actually intend to trade — results on someone else's costs are not transferable.

A forward test on a demo proves the robot executes without errors, which is worth knowing and is not evidence of profit: demo fills have no slippage. A month on a live account at the smallest position size proves considerably more, and costs the price of the lesson.

The honest sequence is strategy tester, then demo, then a live account small enough to be boring, then — only then — the size you intended.

Questions people ask