The advertisement promises a robot with a 95 percent success rate, a one-off payment and profits earned "while you sleep". What follows is painfully repetitive: for two weeks the account grows, the owner adds capital, and a month later less than half of it is left. This is not bad luck or the market's fault — it is the predictable consequence of how such programs are built. Below we explain what an Expert Advisor really is, why off-the-shelf robots fail, and how to test one before you let it near real money.

What exactly is an Expert Advisor?

An Expert Advisor, or EA for short, is a trading program written in MQL4 for MetaTrader 4 or MQL5 for MetaTrader 5. It reads incoming quotes, checks the conditions its author wrote down, and on that basis places orders by itself: opening a position, moving a protective order, closing a trade. There is no magic and no artificial intelligence in it — there is a list of "if, then" conditions, executed without hesitation and without fatigue.

One technical caveat is worth remembering straight away, because it is a source of expensive misunderstandings. An automated system runs only while the platform is open and connected to the broker server. Close the terminal or lose power, and the program stops analysing the market. What does stay on the broker server are the orders already placed: pending orders, stop loss and take profit. Those work even with the platform switched off. That is exactly why people who trade automatically keep the terminal on a rented VPS rather than on a home computer.

What an automated system really gives you, and what it never will

On the benefit side

  • It executes the plan the same way on Monday morning and on Friday evening, without the deviations a human makes out of tiredness or anger after a losing trade.
  • It watches the market during the Asian session as well, when most European traders are asleep.
  • It can be tested on historical data, which means it can be checked before use — something you cannot say about your own intuition.
  • Once configured, it takes little time per day, provided you genuinely supervise it rather than merely switch it on.

On the risk side

  • Off-the-shelf robots from the internet overwhelmingly come with no documented record on a real account — there is only a chart from a historical test, and such a chart can be arranged to taste.
  • The program does not know the market has changed. A system built for a clear trend will keep opening the same positions once the rate settles into a months-long range.
  • Results depend on conditions at your broker: on the spread, on slippage, on connection latency. The same program at two brokers will produce two different outcomes.
  • A purchased robot teaches you nothing. If you do not understand the strategy it executes, you will not recognise the moment it stops working.

Why do off-the-shelf robots stop making money?

The reasons come down to three, and each can be understood without any knowledge of programming.

  1. Overfitting to historical data. The more parameters a program has to set, the easier it is to choose them so that they describe perfectly what has already happened. Such a set of settings describes the past, not the market, and it falls apart on current quotes within weeks.
  2. Optimisation for a pretty sales chart. The author turns the parameters until the equity curve from the test looks like a straight line heading upwards. That is not proof of effectiveness, only proof that somebody spent a long time looking for the best of a thousand variants.
  3. A change in the character of the market. A trend-following strategy earns during periods of clear moves and gives the profit back when the rate enters a range. The program will not notice the difference, because the difference is not among its conditions.

How to test a robot before you let it near your money

  1. A historical test covering several years in the MetaTrader 5 Strategy Tester, and necessarily on real ticks. The choice between Forex Tester and the tester built into the platform depends on your budget and on how precise the data needs to be.
  2. A check on how it handles changed settings. If a small adjustment to one parameter ruins the whole result, the program is fitted to the past. Walk-forward analysis exists for this — a test on data the optimisation never saw.
  3. A test on a demo account over several months, on current quotes. Only here do the differences between a simulation and a real spread come out.
  4. A start at the minimum position size — 0.01 lots — for another few months. This is real money, but the cost of a mistake is symbolic.
  5. An increase in position size only once the live result resembles the result from the test. If the two diverge, there is nothing to increase.

The whole sequence takes six months to a year. Shortening it does not save time; it merely moves the cost of learning from the demo account to the real one.

Where to look for a robot, and where certainly not

The official MQL5 Market run by MetaQuotes is the safest starting point, because products there go through a basic technical check. That technical check is no guarantee of profitability, though — it verifies that a program works, not that it earns.

Purchases from Telegram groups, from forums and from social-media advertisements are a separate category of risk. There is no verification there, no possibility of a refund and nobody to handle a complaint. If the seller does not show a record from a real account covering at least several months, the conversation is pointless regardless of where it is taking place.

Your own robot instead of a bought one

The sensible order is the reverse of the one sellers propose. First you have a strategy that works in manual trading and that you have run with a journal for at least several months. Only then do you write its rules into code, in order to eliminate your own departures from the plan. Automation does not create an edge — it entrenches one, or it replicates the mistake if the mistake was in the strategy.

There are two routes. You can learn MQL5 yourself and start with the simplest robot in MetaTrader 5, or commission the program from a developer, handing over a precise description of the rules. The second route is faster, but it has a catch: if you do not understand the code yourself, the first time the program behaves oddly you will be dependent on somebody else's goodwill.

What to do before you launch your first robot

  1. Describe your strategy on a single page. Write down the entry condition, the exit condition, where the protective order goes and the position size — in sentences unambiguous enough that a developer would not have to guess anything. If you cannot write it down, you have nothing to automate yet, and no purchased program will fix that.
  2. Test one off-the-shelf robot on a demo account for a month. Not in order to buy it, but to see with your own eyes how the result from a historical test diverges from current quotes. It is the cheapest lesson in overfitting available.
  3. Before you buy anything, ask the seller for a record from a real account. What interests you is the number of trades, the deepest drawdown in equity and the period the result covers. A refusal or an evasion is a complete answer and saves you the transfer.

Related material: if you are still weighing up the approach itself, our comparison of algorithmic and manual trading shows what each really costs in time and in skill.