How to Choose a Trading Robot (Without Getting Scammed)

A trading robot, called an Expert Advisor (EA) in the MetaTrader world, is a program that applies fixed rules: entry conditions, position size, stop-loss, exit. On MT5 it runs in the desktop terminal and sends orders to your broker, on your own account.

A robot does not predict the market. It exploits a presumed statistical edge found in past data, which can weaken or disappear. It removes neither losses nor drawdowns (declines in capital). It takes emotion out of execution: useful when the rules are sound, dangerous when they are not.

• It is not an investment product: you remain the account holder and responsible for the risk settings.
• It is not a management service: if someone trades on your behalf, that is portfolio management for third parties, a regulated activity.
• It is not a guarantee: a flattering track record describes the past, nothing more.

For the broader picture, see our guide to algorithmic trading. This guide asks a narrower question: how to assess one specific robot before you buy it or run it.

Our risk disclosure covers these points in detail. Whatever robot you choose, read this kind of document before you pay.

Most dangerous robots can be spotted without reading a line of code: risk management that postpones losses instead of cutting them, marketing that sells a result rather than a method, pressure to buy fast.

Before you pay, check the AMF blacklists and the FINMA warning list. These lists are not exhaustive: a missing name proves nothing, a listed name is reason enough to stop.

The MT5 Strategy Tester report tells you most of what matters about a robot, provided you read the right lines. Ask for the full report, with test settings visible, rather than a lone chart.

Two elements are often missing: real spread and commissions. A very low fixed spread with no commission can turn a losing scalper into a rising curve. Check that the spread is variable (real ticks) and that your account type’s commissions appear in the deal list.

Slippage is harder to simulate: by default the tester executes orders with zero latency, which is kinder than reality. A serious seller explains how they accounted for it.

Finally, look at the period tested: six months of rising markets say nothing about how the robot holds up in a downturn. And ask how many settings were tried. The more parameters were optimised, the greater the risk of a curve fitted to the past.

A robot usually goes through three stages: the backtest, the forward test (in real time, on demo or a small account, over a period the developer could not optimise) and the live account. Results almost always deteriorate from one stage to the next. Up to a point, that is normal.

• Real spread: it widens at session opens, at rollover and around economic releases.
• Slippage: market orders and stops are sometimes filled at a worse price, especially during fast moves.
• Latency: a few dozen milliseconds change a scalper’s results, and matter much less for a long-term system.
• Data: no two brokers have exactly the same quotes.
• Market regime: volatility and market structure evolve.

What is not normal: a robot that is highly profitable in the backtest and clearly loses money as soon as the forward test starts, gaps specific to certain brokers, or a seller who shows nothing beyond the backtest. The faster the strategy, the more sensitive it is to costs and execution.

A simple rule: assume live trading will do worse than the backtest, and check that the robot remains acceptable with higher costs than those simulated. If the edge disappears with a slightly wider spread, it was too thin.

A robot is only reliable if someone maintains it. MetaTrader updates, broker changes, bugs: you need a seller who is identifiable, reachable and still around a year from now.

• Identity: a company or operator name, a physical address, a country and a governing law. Our legal notice shows the minimum you are entitled to expect.
• Public documentation: installation, inputs, general logic and behaviour when losing, available before purchase.
• Support: an identified channel and stated response times. Test it before buying with a precise technical question.
• Refunds: written terms, available before payment. A downloaded digital product is often no longer refundable; that is legitimate, provided it is stated upfront.
• Verified track record: an account tracked by a third party such as Myfxbook, showing “Track Record verified” (the history matches the broker’s) and “Trading Privileges verified” (the publisher has full access to the account). Also look at its length, deposits and withdrawals, and whether it is a live account.
• Regulatory status: a software seller generally needs no licence as long as they never touch your money and send you no personalised recommendations. If they offer to manage your account, they must hold one, verifiable with the regulator.

Be wary too of anonymous testimonials, screenshots of gains in messaging apps and comparison pages that all point to the same affiliate link.

The licence price is rarely the main cost. On an active robot, transaction costs over a year often weigh more.

Do the maths: average cost per trade (spread, commission, estimated slippage) multiplied by monthly trades, plus the VPS. Compare it with the average gain per trade in the backtest. If costs eat most of it, the strategy will not survive a slightly less competitive broker.

A good robot on the wrong account gives a bad result. Check that your setup matches the one the robot was designed and tested for.

• Platform: an MT5 EA (.ex5) does not run on MetaTrader 4, on mobile or on the WebTerminal. It needs a desktop terminal or a VPS.
• Hedging or netting: in hedging mode, several positions coexist on the same symbol; in netting mode, they merge into one. A robot built for one mode can misbehave in the other.
• Symbols: names vary (EURUSD.r, EURUSDm; NAS100, US100, USTEC). The robot must let you choose the symbol or handle suffixes.
• Specifications: contract size, tick value and minimum volume differ, especially on indices and crypto-assets. A lot calculation that does not read them from the broker can open positions ten times too large.
• Trading hours: sessions are expressed in server time, which is specific to each broker.
• Leverage: in the European Union, retail leverage on CFDs is capped (30:1 on major currency pairs, 20:1 on major indices, 2:1 on crypto-assets). A robot tested with more leverage may run short of margin on your account.
• Prop firms: check that EAs are allowed and that the robot’s daily loss limit stays below the firm’s.

Our risk management and configuration pages explain the effect of leverage and how to set sessions in server time. These principles apply to most MT5 robots.

No amount of report reading replaces watching the robot on your own account, with your own broker. Proceed in steps.

1. 1

   Run the backtest yourself

   Run a real-ticks test on your broker, with your costs. EAs on the MQL5 Market offer a free demo that works only in the Strategy Tester.

2. 2

   Install it on a demo account

   Same broker, same account type, same symbols as live. Check lot sizes and that every position has a stop-loss.

3. 3

   Watch it for several weeks

   Long enough to see losses, a quiet period and an economic release. Compare spread, slippage and trade frequency with the backtest.

4. 4

   Go live with reduced risk

   Small capital or a lower risk per trade than your target. The goal is to validate execution, not to make money.

5. 5

   Set stop criteria in advance

   Decide before you start which drawdown will make you stop. Written down with a cool head, the rule is easier to keep.

6. 6

   Scale up in steps

   Only after a period that matches expectations, never to recover a loss.

For the technical installation (MQL5\Experts folder, permissions, Experts tab), follow our installation guide, which broadly applies to most MT5 EAs.

If you cannot tick most of these boxes, wait, ask the seller for clarification or walk away.

• Every position has a stop-loss placed with the broker from the moment it opens.
• No martingale: size never increases after a loss.
• No grid: no position is added against a losing position.
• The backtest report is complete, on real ticks, with variable spread and commissions.
• Equity drawdown is shown and consistent with balance drawdown.
• The number of trades is sufficient and the period covers several market regimes.
• A forward test or a live track record verified by a third party is available.
• No promise of monthly returns, no “guaranteed”, no “risk-free”.
• The seller is identifiable: legal notice, address, reachable support.
• The documentation is public and the risk settings are yours to set.
• Licence and refund terms are written down before payment.
• The robot is compatible with your broker, account type and leverage, and you have tested it on demo.

Algo Market Labs® designs and sells Expert Advisors for MetaTrader 5. Publishing this checklist without applying it to our own products would be inconsistent. Here is what you can check.

• Stop-loss: every order is sent with a stop-loss and a take-profit; the code refuses any order without both.
• No grid, no martingale: one position at a time, size is never increased after a loss and no position is added against a losing one. Principles and methodology: strategy page.
• Testing: simulated backtests on data from an MT5 broker (standard account, no commission), with variable spread and swaps, without simulated slippage. Real ticks are available from June 2025 only; before that, MetaTrader reconstructs ticks from 1-minute bars. Settings are chosen on data up to 31 December 2024, then validated out of sample from 1 January 2025 to 26 September 2026 against criteria set in advance (details, and one disclosed exception for APEX, on the strategy page). A demo account follow-up on Myfxbook is planned but has not started yet: no live or demo track record is available to date.
• Risk: you set the risk per trade and the maximum daily loss; once that threshold is reached, the EA opens no new position.
• Capital: it depends on the minimum position of your account at your broker, not on the EA. With the minimum positions of our test account, NOVA takes its signals from 500 USD at its default setting (2% risk per trade), APEX from 500 USD with its Dynamic profile (higher risk), QUANTIS from 2,000 USD at 4% risk per trade and VECTOR X1 from 3,000 USD at level 5. With less, the EA skips signals instead of enlarging the position, and says so in its log. On a small account, the licence price can exceed the possible gains. Details: capital and minimum position.
• Public documentation, including the frequently asked questions, available without buying.
• Licence: one year or for life, no automatic billing, for two MT5 accounts (demo or live). The right of withdrawal ends once the download has started, as our terms state before payment. Test the EA on a demo account before going live.
• Identifiable seller: a Swiss sole proprietorship based in Sierre, with support 7 days a week by email and on Telegram.

Our four systems have distinct profiles. NOVA follows the trend of Ethereum (ETHUSD), buy only, with few positions and an aggressive profile; it is built for accounts from 500 USD. APEX trades intraday momentum on the Nasdaq 100, buy only and without holding positions overnight, with an aggressive profile. QUANTIS trades Bitcoin on the buy side only, over the long term, with a moderate profile. VECTOR X1 is a trend-following EA on gold (XAUUSD), buy only, that opens few positions, with an aggressive profile adjustable from 1 to 5. All four send a stop-loss and a take-profit with every order, hold one position at a time and use neither a grid nor a martingale. None of them suits everyone and none guarantees a result: apply the same checklist to them as to any other robot.