Three points to keep
- An expert advisor (EA) is a program that runs inside a trading platform such as MetaTrader and places or manages orders by fixed rules.
- Backtests can show whether rules were consistent in the past; they cannot prove future profit, and are easy to overfit.
- Costs, slippage and changing conditions often erase a small edge, so treat any EA sold with performance promises as a warning sign.
EA trading is automated trading inside a retail platform: an expert advisor watches prices and places, modifies or closes orders according to rules written in code. The term comes from the MetaTrader platforms, which allow programs written in their own languages to run alongside the charts. The same idea exists on other platforms and in crypto, where it is usually called a trading bot. Automation removes hesitation and fatigue; it does not create an edge that the rules do not already have.
How an EA works
An EA is triggered each time a new price arrives (a "tick") or at set intervals. It reads the data it needs, such as prices, indicator values and open positions, then checks its rules. If conditions are met, it sends an order to the broker's server with a size, a stop-loss and a target. Some EAs only manage trades a person opened, for example trailing a stop.
Typical rule families include trend-following (buy after a breakout, sell after a breakdown), mean-reversion (fade moves away from an average), grid strategies (layer orders at fixed intervals) and martingale systems (increase size after a loss). The last two can show smooth equity curves for long periods and then lose most of the account in one sustained move, because they add to losing positions.
Backtests and what they can prove
A backtest runs the EA on historical prices to see how its rules would have behaved. It is useful for catching coding errors and understanding how often the rules trade. It is weak evidence of future profit, for several reasons:
- Overfitting: tuning parameters until the past looks good produces rules that fit noise.
- Data quality: backtests on minute bars rather than tick data can misjudge where stops and targets filled.
- Costs: a fixed spread assumption hides the wider spreads of news and rollover periods.
- Regime change: a strategy that worked when volatility was low may fail when it rises.
An illustrative expectancy check
Suppose an EA wins 55% of trades with an average win of $80 and loses 45% with an average loss of $90. Expected result per trade = 0.55 × $80 − 0.45 × $90 = $44 − $40.50 = $3.50. If spread, commission and slippage add $4 per trade, the expectancy becomes $3.50 − $4 = −$0.50. An edge that small disappears in live trading.
Running an EA: platform, broker and VPS
An EA only works while the platform is open and connected. Many users rent a virtual private server (VPS), a remote computer that runs continuously, often located near the broker's servers to cut delay. A VPS solves uptime; it does not improve a strategy. Check that the broker permits automated trading, how it treats very short holding periods, and whether its execution model and spreads match what the backtest assumed.
Demo testing on live prices (forward testing) for weeks before using real money shows how the EA handles real spreads and fills, though demo execution can still be kinder than live. See what demo accounts can and cannot show.
Keep a log of every change you make to an EA's settings, with the date. Without it, you cannot tell whether a change in results came from the market or from you.
Bots in crypto markets
Crypto exchanges offer grid bots, dollar-cost-averaging bots and API access for custom programs. A bot can follow any market, whether bitcoin or the algo coin price of Algorand's ALGO token, but it cannot know where that price is heading; it only executes rules. Two risks are specific to crypto: API keys with withdrawal rights enabled can be abused if leaked, and markets run around the clock, so a malfunction can go unnoticed for hours. Create keys with trading permission only and restrict them to known IP addresses where the exchange allows it.
A sensible testing sequence
- Read or document the rules in plain language, including how size is chosen and what happens after losses.
- Backtest on good-quality data with realistic spreads and commissions, and look at the worst drawdown, not only the final return.
- Test on data the rules were not tuned on (out-of-sample).
- Forward-test on a demo account for long enough to see several market conditions.
- If you go live, start with the minimum size and compare live fills with demo fills.
Judging EA trading products
EAs and "signal bots" are sold widely, often with screenshots of steady gains. Apply the same tests as to any investment claim:
- Are results verified on a live account over a long period, or only a backtest or demo?
- Does the seller explain the rules, or only the returns?
- Does the equity curve look too smooth? Grid and martingale systems often do, until they fail.
- Is the product tied to a specific broker the seller is paid to refer you to?
- Are returns described as guaranteed or fixed? That alone is disqualifying.
Many trading scams now present themselves as AI or bot services. The warning signs are the same as in our scam guide.
Reader questions
Do I need to know how to code to use an EA?
Not to run one, but without reading the code or a clear description of its rules you cannot know what it will do in unusual conditions. Understanding the logic matters more than writing it.
Is EA trading allowed by all brokers?
Many brokers allow it, but some restrict certain behaviours such as very high-frequency order placement or specific arbitrage methods. Check the terms before running any automated system.
Can an EA guarantee profits?
No. An EA follows rules; whether the rules make money depends on markets and costs. Any claim of guaranteed returns from a trading program is a warning sign.