Trading bots · Reliability

Are trading bots reliable? How to check before you pay

As a category, the question has no answer: "trading bot" is not a level of quality, it is a way of executing. Some bots do exactly what they promise and some do not survive their first month outside the backtest, and from the outside the two look much alike.

What you can check is one particular bot, before you pay for it. This is that checklist: what a bot can and cannot do, why an excellent backtest proves almost nothing, what a history being "verified" actually means, and the five warning signs worth looking for. None of them promises to be right. All of them can be checked in an afternoon.

Sources read on 28 August 2026 · linked at the foot of the page

This page does not assess any particular bot

What a trading bot is and what it cannot do

The real advantage of automating is discipline, not accuracy.

A trading bot is a program that executes a set of predefined rules on a platform without anybody pressing the button. It does not predict the market, it does not guarantee a result, and being automated does not take it out of the risk.

What it does well is execute: at any hour, without hesitating, and without drifting from its own rules because a bad run has started. That is a genuine advantage and it is worth not underrating, because most of the rules a human trader breaks get broken at precisely the moment it costs most. But it is an advantage over execution, not over accuracy: a bot only knows what was written into the rules it was given.

Nor does it change the risk of the instrument it trades. When ESMA restricted the marketing of CFDs to retail clients in 2018, it justified the measure with the national regulators' own analyses, which showed that between 74% and 89% of retail accounts lose money, with average losses per client between €1,600 and €29,000. That is the starting point of the product. Automating the execution does not move that figure in any known direction, for better or worse.

And it does not get around the rules of the account it trades in. If the bot runs in a funded account, the prop firm's rulebook applies to it exactly as it applies to a person, with the added problem that a program cannot tell it is breaking the rules. That has its own section further down.

Why an excellent backtest says almost nothing

A spectacular backtest is cheap to produce, and that is precisely the point.

Because trying many combinations against the same history eventually produces a good-looking curve by pure chance. Bailey, Borwein, López de Prado and Zhu proved it formally in 2014 in the Notices of the American Mathematical Society: "high simulated performance is easily achievable after backtesting a relatively small number of alternative strategy configurations".

The important words in that sentence are "relatively small". This does not take a laboratory or thousands of runs: a modest number of alternative configurations is already enough to turn up something that looks excellent.

The backtest is the result of the search, not the evidence
Nobody publishes how many configurations they tried before settling on the one they show you. That number is exactly what you would need in order to estimate how much of the curve is a finding and how much is fitting, and the authors point to it as the central problem: because the number of attempts goes unreported, an investor cannot evaluate the degree of overfitting in what they are being shown.
A tuned strategy inherits the stretch it was tuned on
Fitting to one particular period drags that period's conditions along with it: its volatility, its spreads, its trading hours, which instruments were moving and which were not. When the market stops resembling that stretch, the execution is still correct but the edge is no longer there. The finer the fit, the narrower the range of conditions in which it is any use.
The only thing that means anything is what happens out of sample
That is, the behaviour on data the strategy did not see while it was being built, and better still in live trading, where slippage, commissions and opening gaps exist. A backtest has neither of those things: by definition it runs on what already happened, and under ideal conditions.

None of this makes the backtest useless. It is good for ruling things out: a strategy that does not even work on the past is not going to work on the future. What it cannot do is demonstrate the opposite, and that is why an excellent backtest is not an argument, it is a starting point.

What "verified track record" actually means

The badge says the figures are the broker's. It says nothing else, and that is where the misunderstanding lives.

It means one thing only: that the figures being published match the ones the broker holds. It does not mean the strategy is any good, nor that it will carry on working, nor — depending on the platform and the level — that the account is trading real money.

Myfxbook splits it into two verifications that are worth not confusing. The one covering the history says that "the account's trading history shown on Myfxbook matches the account's trading history provided by the broker on the trading platform". The one covering trading privileges says something different: that "the individual who added the trading account possesses the master password, signifying ownership or the ability to execute trades on the account". The first validates the numbers; the second validates who registered the account. An account can have one without the other.

FX Blue is more explicit still, and its documentation is the clearest description in the industry of what cannot be known. It distinguishes three levels:

LevelWhat it guarantees

VerifiedVerified

Where the data comes from: The broker, by account sync
What it confirms: The trades come from the broker
Open to tampering: Not by the user

Semi-verifiedSemi-verified

Where the data comes from: Publisher EA with anti-tamper controls
What it confirms: Results almost certainly genuine
Open to tampering: In theory yes

FX Blue puts it like this: results are "almost certainly genuine", but "it is theoretically possible for users to manipulate the data which is sent to FX Blue".

UnverifiedUnverified

Where the data comes from: Reports uploaded by FTP, or backtests
What it confirms: Nothing checkable
Open to tampering: No way of knowing

And here is the sentence most worth remembering: "It is impossible to verify whether the results have been manipulated, or whether they are from a demo account or a real one".

Out of that come the two questions to put to any screenshot of results. First: which level is this account at? A screen capture, a PDF or a hand-uploaded report sits in the last one, where by definition a real account cannot be told apart from a demo. Second: even where it is verified, can you see the bad stretch? A verified history of which only the good part is shown is still half a history.

Put the other way round, "verified" is not a synonym for "good". It is a synonym for "checkable", which is a precondition and not a conclusion.

The five warning signs

None of them proves fraud on its own. What matters is not the sign, it is the one that gets no answer.

Guaranteed, fixed or "risk-free" returns
The CNMV, the Spanish securities regulator, sets this out as grounds for suspicion without qualification in its guide to unauthorised investment firms: promises of exceptional returns without risk should make us distrust the offer immediately. And on the financial education site run by the CNMV and the Bank of Spain, the same idea in other words: if you are offered very high returns with no risk attached, be careful, because that does not exist. That guide is written about firms providing investment services without authorisation, which is not the same as selling software, but the grounds for suspicion carry over untouched: a fixed monthly percentage promised on something that trades in a market is this same thing applied to bots.
A curve with no downward stretch
Every system that trades in a market has bad periods. If none of them shows up in the sales material, what you are looking at is not a system without losses: it is a selection. Ask for the worst month and for the longest run of consecutive losses. Those are two numbers anybody with a real history gives from memory, and whoever has not got one gives themselves away in the answer sooner than in the figure.
Results that cannot be audited
Screenshots, a PDF, a hand-uploaded report, a video of the platform. None of that is verification in the sense of the previous section: there is no way to know whether it has been tampered with or whether the account was real. What is checkable is a link to a verified account on Myfxbook or FX Blue, with the full history and the verification level in plain sight. If that link does not exist, the right question is not why the results are so good, it is why they are not where they can be checked.
Testimonials that cannot be corroborated
No name, no date, no account behind them and no way of reaching the person. A testimonial like that carries no information, because writing one costs the same as writing twenty. The ones worth something are the ones you can follow back to somebody identifiable who is willing to answer, and they are considerably rarer.
No company identification
This is the easiest one to check and the one most people skip. Spain's Ley 34/2002 requires anybody providing services over the internet to make their registered name, their address and a means of direct and effective contact available permanently, easily, directly and free of charge. If the site carries no registered name, no tax number and no address, the practical question stops being whether the bot is any good and becomes who you would claim against if it is not. It is also worth running the name through the CNMV's public warnings search, bearing in mind what the CNMV itself points out: a firm that has not been the subject of a warning is not necessarily an authorised firm, it may simply be that its activity has not been detected yet.

These five are not here as a sales argument. They are the same five to look at any bot with, including the system sold on this site. Asking costs nothing, and anybody who has the answers should not mind giving them.

What prop firms say about bots

They allow them. What they limit is not automating, it is the activity the bot generates.

A bot is not banned in a funded account, and that is worth saying because it is a common worry. FTMO accepts expert advisors both in the challenge and in the funded account; its restrictions fall on what the bot does, not on the fact that a program is the one doing it.

The most concrete limit is on activity: FTMO treats an account as hyperactive when it makes "more than 2,000 server requests per day on individual simulated trades or pending orders". Pending orders count there too, not only closed trades, so a bot that adjusts its stop frequently hits that ceiling far sooner than its number of trades suggests.

And there is one point that lands squarely on anybody buying a ready-made bot: FTMO warns that a third-party EA may be running for other traders at the same time, which exposes the buyer to its maximum capital allocation rule. The full breakdown, with the five figures and their official source, is on the page about FTMO and expert advisors.

For the purposes of this page, what matters is that "allowed" is not "without conditions", and that complying with the rulebook is the account holder's responsibility. A bot that makes you break a rule does not give you the challenge back.

Sources · read on 28 August 2026

Frequently asked questions

Are trading bots reliable?

As a category the question has no answer, because "trading bot" describes a way of executing rather than a level of quality. What can be answered is whether one particular bot is checkable, and that comes down to four things: whether its track record is verified by a third party and at which level, whether the worst stretch is shown and not only the best, whether the seller is identified by registered name and address, and whether it promises a result. A bot that fails all four is not necessarily a fraud, but there is no way to tell it apart from one.

Does a good backtest mean the bot works?

No. It means somebody found a combination of rules and parameters that would have worked on one particular history. Bailey, Borwein, López de Prado and Zhu proved in 2014 that testing a relatively small number of alternative configurations is enough to produce high simulated performance, and that because almost nobody reports how many were tried, the buyer cannot estimate how much of that curve is overfitting. A backtest is useful for ruling a strategy out, not for confirming it.

What does a verified track record actually prove?

That the published figures match the ones the broker holds, and nothing else. Myfxbook defines that verification as the trading history shown on its site matching the history the broker provides on the trading platform, and keeps it separate from trading privileges verification, which confirms a different thing: that whoever added the account holds the master password. Neither of the two says whether the strategy is any good or whether it will carry on working.

If the bot loses money, who answers for it?

The account holder, who takes the losses: the money sits in their account and the trades are executed there. Buying a bot means buying software and its installation, not hiring somebody to manage capital, which is why a trading loss is not covered by the seller. Before paying, get it in writing: exactly what you are buying, whether there is a refund and on what grounds, and which company stands behind that contract.

Why does a bot that worked last year stop working?

Because its edge was tied to particular market conditions, and those conditions change. A strategy tuned on one stretch of history inherits that stretch: its volatility, its spreads, its trading hours. When the market stops resembling it, the execution is still correct but the edge has gone. The finer the fit to one period, the narrower the range of conditions in which it is worth anything.

What should I ask somebody who wants to sell me a bot?

The CNMV, the Spanish securities regulator, publishes a list of questions for financial offers that carries over almost intact: how long they have been in this business and what their professional experience is, which company is behind it and whether it is registered, what the risks are and how much you can lose, what fees you pay and how they are calculated, and whether they can send you the documents and the contract in writing so you can read them calmly. Add two that are specific to a bot: ask to see the verified track record with the worst month included, and ask how many other people they have sold the same system to. The CNMV itself warns that asking is not enough, because people who set out to defraud are trained to sound persuasive and will dodge the answers with reasonable-sounding arguments, so what has to be checked is the answer.

Continue reading

Three more guides on this, or on what sits next to it.