The equity curve looked perfect: twelve months of smooth, steady growth, barely a red month, the kind of screenshot that sells. Then one Tuesday the market trended hard for three days and the account was gone — every gain, plus the deposit. That shape — smooth, then cliff — is the signature of a martingale system, and learning to recognize it is the first thing anyone should know before trusting their money to a forex Expert Advisor.
An EA — Expert Advisor — is an automated trading program that runs on MetaTrader 4 or 5, written in MQL4/MQL5. You give it rules: if a fast moving average crosses above a slow one, buy; if price hits this level, exit; risk 1% per trade. The EA monitors the market and executes without you, 24 hours a day, five days a week, never tired, never emotional, never breaking its own rules out of fear or greed. Those are real advantages: discipline is precisely where human traders fail, and a robot connected to a VPS near the broker's servers catches signals at 3 AM that a sleeping human would miss.
But notice what an EA does not do: it does not create an edge that the rules lack. Automation multiplies whatever the underlying strategy contains — including nothing. The industry data makes the point brutally: European regulators report 74–89% of retail CFD accounts lose money, and an EA does not change that number; it only automates the decisions that produce it.
The market where EAs are sold makes this worse. The MQL5 Market and third-party vendors list hundreds of products promising 10–20% monthly returns, and the overwhelming majority either die on contact with live markets or run strategies that are statistically guaranteed to blow up eventually. The standard failure modes are worth memorizing. Curve-fitting: parameters tweaked until the backtest looks perfect — a spectacular backtest is famously the easiest thing to fake in trading. Tick and spread sensitivity: the EA works on broker A's data feed but not broker B's. Fabricated or cherry-picked track records: demo accounts presented as live, or multiple accounts where only the winner is shown. And the smooth-then-cliff equity curve of martingale systems that double position size after every loss, or grid systems that pile up positions without stop-losses — both show months of charming growth followed by a single adverse trend that erases everything, because doubling down is mathematically guaranteed to meet a streak it cannot survive.
The real-world cautionary tale is serious money, not forum folklore. Mirror Trading International marketed automated bot trading with claims of roughly 10% monthly returns; the US CFTC later called it a $1.7 billion Ponzi scheme in which little actual trading occurred, with fake accounts built on MetaTrader demo balances, and ordered its CEO to pay over $3.4 billion. Regulators including the FCA and CFTC regularly warn about "AI trading" apps with fake celebrity endorsements and promised daily profits, where the "robot" is just a lead-capture page feeding call centers. The pattern to refuse: guaranteed returns, sign-up funnels, offshore brokers with no name, and AI as a marketing word. Real AI in trading exists — institutions use it for research and execution — but no licensed product guarantees profits, ever.
Separating the rare legitimate product from the noise is a checklist, and it eliminates the vast majority of bad options before any money moves. Demand a live, independently verified track record — Myfxbook or FXBlue, minimum twelve months, ideally twenty-four; demo accounts do not count. Check the maximum drawdown, not just the return; if it is not published, walk away. Verify at least two hundred trades, because anything less is statistical noise. Inspect the equity curve shape — smooth with occasional dips is fine; smooth-then-cliff is a martingale. Read the strategy description: if the seller cannot explain why the system makes money and under what conditions it stops, there is no edge to evaluate. Search independent forums for complaint patterns. Then forward-test on demo for one to three months on the broker you will actually use, and go live at minimum size before scaling anything.
The honest alternative, and the one experienced traders actually recommend, is building your own EA — even a simple one. Writing it means you know every rule, can diagnose why it lost money on a specific day, understand its assumptions, and can recognize when market conditions have moved outside the logic. A trader who does not understand their EA's logic cannot evaluate its performance or know when to stop it, which makes every automated trade an act of faith. MQL5 is accessible with basic programming knowledge, and Python routes through broker APIs exist for those who prefer it.
And even a well-built, well-tested EA is not "set and forget." Market regimes change; a trend-following system prints money until consolidation arrives, and no code understands the change unless a human anticipated it. Set an account-level hard limit — say, disable the robot at 20% drawdown — and monitor it like the employee it is, not the oracle the ads promise.
The summary in one sentence: an EA is a discipline machine, and discipline applied to a strategy with no edge produces losses with perfect consistency. The robot never gets tired, never gets scared, and never gets lucky — it just executes, forever, whatever you actually gave it. Vet the logic before you lend it the keys; the smooth equity curve that ends in a cliff is always someone else's screenshot until it becomes your account.