Do Expert Advisors actually work?
An EA is software. It works exactly as well as the rules inside it and the testing behind it. Here is why many bought EAs disappoint, what to check before you trust one, and when a custom EA makes more sense.
Based on our team’s research and live testing since 2018.
The short answer: an Expert Advisor does what it was built to do. It follows rules exactly, around the clock. Whether that makes money depends on the rules, the trading costs and the market. The EA is the tool, not the edge.
We do not have reliable data on what share of commercial EAs lose money, and nobody selling one does either. What we can explain is how they fail. The mechanisms are well understood, and each one can be checked before you risk money.
What does an EA actually do?
An EA is a program that runs inside MetaTrader. It reads prices, checks its rules, and opens, manages and closes trades. It does not predict the market and it does not learn by itself unless it was built to. If the rules have a real edge after costs, an EA can trade that edge more consistently than a person can. If they do not, the EA loses more consistently. Automation magnifies the strategy, good or bad.
Why do so many bought EAs fail live?
1. They were fitted to the past
Give an optimiser enough parameters and it will find settings that made money on any stretch of history. Those settings describe the past noise, not a pattern that repeats. The tell-tale sign is a near-perfect backtest that falls apart on data it was not tuned on. See how to avoid overfitting.
2. The costs were left out
A backtest with a fixed low spread and no commission flatters every strategy, and it flatters scalpers most. If a strategy aims for a few pips per trade, a pip of cost on each round trip can remove most of its profit. Live, the spread also widens at session opens and around news, which is often when these EAs trade.
3. The market changed
Trend systems lose in sideways markets. Range systems lose in trends. Volatility, spreads and interest rates shift over the years. An EA that is not expected to fail in some conditions has usually not been tested in enough of them.
4. The equity curve is marketing
A sales page chooses its period, its symbol, its broker and its settings. It may show a backtest as if it were live trading, or a demo account as if it were real money. None of that is evidence of how the EA will trade on your account from today.
5. Your setup is not the seller’s setup
Symbol suffixes, minimum stop distances, filling modes, server time and account type differ between brokers. An EA built for one broker can behave differently, or fail to place orders, at another. The usual causes are listed in why an EA works in the backtest but not live.
6. You cannot see inside it
With only a compiled file, you cannot tell whether a losing month is normal for the strategy or a sign that something is broken. You also cannot fix it when your broker changes something.
Why can a smooth equity curve hide the biggest risk?
Some of the smoothest curves come from martingale systems, which double the position after each loss so that one win recovers everything. The curve rises in small, steady steps, until a long losing streak arrives. Here is the arithmetic, with illustrative numbers: a USD account on EURUSD, where 0.01 lot is worth about $0.10 per pip, a 20-pip stop and a 20-pip target, and a start at 0.01 lot.
| Losses in a row | Lot size on that trade | Loss on that trade | Total loss so far |
|---|---|---|---|
| 1 | 0.01 | $2 | $2 |
| 4 | 0.08 | $16 | $30 |
| 6 | 0.32 | $64 | $126 |
| 8 | 1.28 | $256 | $510 |
| 10 | 5.12 | $1,024 | $2,046 |
Each sequence that ends in a win earns about $2 before costs. On a $1,000 account, nine losses in a row take more than the whole balance, and the margin needed for the large lots may run out even sooner. A backtest of a year with no long streak shows a beautiful line. It tells you nothing about the streak that has not happened yet. Grid systems that add positions without a cap carry a similar risk; see grid and martingale EA risks.
How can you judge an EA before you trust it?
| Check | What a good sign looks like | Red flag |
|---|---|---|
| Track record | A real-money account, verified by a third party, long enough to include losing periods | Only backtest screenshots, a demo account, or a few good weeks |
| Your own test | You can run it in the Strategy Tester on your broker’s real ticks | The seller avoids letting you test it |
| Costs | The report shows spread and commission on every trade | A fixed, low spread and no commission |
| Risk per trade | A stop on every trade and size calculated from it | No stop, lot multipliers, an uncapped grid |
| Unseen data | Settings fixed first, then tested on a later period | Settings re-optimised every few weeks |
| Claims | Explains the logic and when it is expected to lose | “Guaranteed”, “never loses”, a near-perfect win rate |
On the MQL5 Market, the demo version of an EA can be run only in the Strategy Tester, not on a live chart. That is enough to test it yourself. Use “Every tick based on real ticks”, your broker’s account, and a period the seller did not show you. The setup is in how to backtest with real tick data. Then check the lot sizes it uses with the position size calculator, so you know what each trade really risks.
When does a custom EA make sense?
A custom EA makes sense when you already have a method you understand, can write it as exact rules, and want it executed consistently: around the clock, without hesitation, or within a prop firm’s limits. Compared with a bought robot, you get three things:
- You know the logic, so you know when a losing period is expected and when something is wrong.
- You own the source code, so it can be changed when your broker or your firm changes something.
- The risk rules are yours: stops, daily limits and position sizing set to your account, not to a sales page.
A custom EA does not turn a weak idea into a strong one. Before we build, Harshal K. and Bilal M. look at the strategy’s rules and any test results against predefined criteria, and we tell you if the testing does not support the idea. If it does, I build it as a custom MT4/MT5 Expert Advisor. If you already own an EA, we can test it independently or fix the problems in it.
So, are forex robots profitable?
Some traders run automated strategies profitably. When they do, the reason is the strategy, the cost control and the risk limits, not the fact that a robot is placing the orders. Anyone who promises a robot will be profitable is promising something nobody can know.
Checklist before you trust an EA
- Tested by you, on your broker’s real ticks, with full costs
- Tested on a period it was not tuned on
- A stop on every trade; no lot multiplication after losses
- Drawdown and losing streaks you could live through
- A track record on real money, or a demo run of your own
- Started live at the smallest size
Quick answers
Are forex robots a scam?
Not all of them. An EA is just software that follows rules. The problem is usually the marketing: backtests shown as proof, hand-picked periods, and promises that no honest seller can make.
Can an EA make money on its own?
An EA can only follow its rules. If the rules have an edge after costs, it can capture it consistently. If they do not, it will lose consistently. It still needs monitoring either way.
What is the safest kind of EA?
There is no safe EA, but some designs are less fragile: a stop loss on every trade, position size calculated from the stop, no lot multiplication after losses, and hard daily and total loss limits.
Should I trust a high win rate?
Not on its own. A 90% win rate with rare, very large losses can lose money overall. Look at the average loss, the worst drawdown and the losing streaks together.
Can I use a bought EA on a prop-firm account?
Some firms restrict third-party EAs that many traders run at once. Check your firm’s current rules before you start an evaluation.