Do Forex Expert Advisors Really Work? What the Evidence Says, and How to Test One Yourself

The short answer: two different questions

People ask whether EAs work and mean two things. The first is whether the software does what it says: opens a trade when the rule fires, places the stop, closes at the target. A competent EA does this reliably, and it is the whole point of automation. The second is whether following those rules makes money after spread, commission, swap and slippage, in the market you will actually trade, for longer than the period it was tuned on. That is much harder, and automation does nothing to make it easier.

An EA is a way of executing a trading idea consistently. If the idea has a real edge, the EA applies it without the emotional mistakes a person makes. If the idea has no edge, the EA applies that consistently too, and loses consistently. Read what an Expert Advisor does in MetaTrader 5 if you want the mechanics first, and come back here for the question of evidence.

74-89%

of retail CFD accounts typically lose money (national regulator analyses cited by ESMA, 2018)

30:1

maximum EU retail leverage on major currency pairs; gold is limited to 20:1

50%

of required margin is the level at which EU brokers must close retail positions

Be careful with the first figure. It describes retail trading on leveraged products as a whole, mostly done by hand, and says nothing specific about EAs. It matters here for a simple reason: it is the starting point. A buyer who assumes an EA changes those odds without checking has not earned that assumption. The guide to EA trading under ESMA rules explains the limits behind the other two figures.

What an EA can and cannot do

The right column is where most disappointment comes from. The CFTC puts it plainly in its advisory on trading bots: "AI technology can't predict the future". The same applies to any rule-based system, however advanced the marketing sounds.

Five reasons most EAs disappoint

The first reason deserves a closer look because it is the most common and the least visible. Bailey, Borwein, López de Prado and Zhu describe how testing enough variations of a strategy on the same history will almost always turn up one that looks excellent by chance. The more variations tried, the better the best one looks, and the less it means. Our guide to optimising an EA without overfitting shows how to check for it.

What the regulators say about trading bots

The US Commodity Futures Trading Commission has published a customer advisory titled "AI Won't Turn Trading Bots into Money Machines". It is written about fraud, but the patterns it lists are the same ones a careful buyer should look for in any bot marketing.

• Promises of unreasonably high returns, or returns presented as assured, and claims of gains in the tens of thousands of percent or "100 percent 'win' rates".
• Marketing that says no trading experience is needed.
• Referral bonuses and affiliate payouts that reward bringing in new buyers.
• Fake customer accounts and balances created on MetaTrader demo accounts.
• Promotion by social media influencers and strangers met online.

One case in the advisory involved a promise of at least 10 percent a month. The CFTC points out that this is more than 200 percent a year. That is a useful yardstick for any advertised monthly figure: multiply it by twelve and ask whether any honest, long-lived strategy could sustain it. The fourth item is the one most relevant to MetaTrader users, because a demo account can show any balance the creator likes. That is why the guide to verifying an EA track record insists on records the seller cannot edit.

What counts as evidence, from strongest to weakest

The four grades are the same ones used in our EA Verdicts, where each reviewed EA is scored on eight published criteria. Evidence quality carries the largest weight, because without it the other seven are guesses. You can read the full scoring method, or filter every scored EA by design, risk label and score in the EA screener.

A worked example: the same EA at three evidence levels

Numbers on their own mean little until you see them next to what stands behind them. The table below scores three hypothetical EA profiles with the EA strategy check, which counts red flags in a backtest or live record. None of these is a real product. They exist to show how the same headline numbers change meaning as the evidence behind them changes.

Profile C is the one that looks best on an advertisement: a 91% win rate, a drawdown of 8% and 310% growth. It is also the profile with the most problems. Each average win of 4 is set against an average loss of 38, so a single bad run erases months of gains. The test used open prices and no costs, the sample is small, and a grid with no hard stop has not yet met a market that breaks it. Profile A looks plain by comparison and scores better, because the evidence behind it is sounder. Profile B is the same plain numbers with proof, which is what you are looking for.

A four-week test before you risk real money

1. 1

   Step 1: Backtest on real ticks, with costs

   Run the exact settings on real tick data with your broker's spread and commission. The Strategy Tester guide walks through the settings, and the backtest report guide shows what to read first.

2. 2

   Step 2: Check it on data it has not seen

   Hold back a period the optimiser never touched, or run a walk-forward analysis. A strategy that only works on the data it was tuned on has not shown an edge.

3. 3

   Step 3: Forward test on a demo account for at least four weeks

   Use your own broker and the same settings. The first 30 days plan sets out what to watch each week.

4. 4

   Step 4: Start live small, with a loss limit written in advance

   Size the account so the tested maximum drawdown is a loss you can accept. The account size guide has worked examples, and the drawdown guide covers what to do when it arrives.

5. 5

   Step 5: Compare live with the backtest of the same week

   A small, steady gap from costs is normal. A large or growing one has a cause you can find, as the guide on why live results differ from backtests explains.

So, do forex EAs work?

Some do, for a while, in the conditions they were built for. Many do not, and the ones that fail often looked the best on paper. An EA that works is not a mystery product. It is a rule set with a real edge, risk that is bounded and shown, a record you can verify, and a size that suits the account. Take away any one of those and the odds move against you.

The practical stance is neither belief nor dismissal. Treat every EA, including the ones we sell, as a claim to be tested. We sell EAs, so weigh what we say accordingly, which is why we publish the scoring method and the red flags we look for. Use the EA buying checklist before you pay anyone, and the lot size calculator to turn a risk limit into a lot size. An EA that survives that process has earned a small live start, nothing more.