What are you actually buying when you buy an Expert Advisor?
A backtest is not a track record. It is a description of what would have happened, written by someone who already knew what happened.
Why do EA sales pages all look the same?
Because they are all solving the same problem: proving something works without any live evidence that it does. The available material is a backtest, a screenshot of an equity curve, and some testimonials. So that is what every page contains, in roughly the same order, with roughly the same claims about drawdown.
The uniformity is itself informative.
When an entire market’s marketing converges on one format, it is usually because the format is the only one the underlying product supports. A strategy with five years of audited live results would lead with them.
What is wrong with a backtest?
Nothing, as a research tool. Everything, as evidence. A backtest is generated by running rules over data the author already has, and the author can adjust the rules until the output looks good. That process has a name, curve fitting, and it produces beautiful historical results with no predictive content whatsoever.
The tell is precision. A strategy that trades on a 47-period moving average with a 2.3% stop and a Tuesday filter has almost certainly been tuned until those exact numbers worked. Round numbers are more honest than specific ones, which is the opposite of the intuition most buyers bring.
| What the sales page shows | What it proves | What to ask for instead |
| Backtest equity curve | the rules fit past data | out-of-sample results |
| Screenshot of profits | one account, one period | a full statement, all trades |
| Percentage win rate | very little on its own | average win against average loss |
| Maximum drawdown | the worst that already happened | drawdown in live trading |
| Years of development | effort, not edge | when it started trading live |
| Testimonials | some buyers were happy early | results after twelve months |
Six standard claims and what each is worth. None of them is fraudulent. All of them answer an easier question than the one a buyer is actually asking.
Does the spread assumption matter that much?
Enormously, and it is where most backtests quietly fail. A strategy tested at a fixed spread has been measured against a market that never existed.
Live testing on a funded account measured EUR/USD in July 2026 at roughly 0.1 pips through the London session, and 0.9 pips during a US payrolls release, on the same account on the same day. For a strategy taking a handful of pips per trade, that difference is the entire margin.
Run it at a constant 0.1 and it prints money. Run it at the real distribution, with the wide readings clustered exactly around the news events that generate the moves it trades, and it frequently does not. Very few vendor backtests model this.
What about the ones that publish live results?
Better, and worth taking seriously, with two checks.
First, is the live account real money or a demo, because demo fills are not real fills and a demo track record proves considerably less than it appears to. Second, how many strategies did the vendor launch in the same period, because publishing the survivor from ten attempts is selection, not performance.
That second point is the one nobody asks about. A vendor running ten variants and promoting whichever survived has produced an impressive chart through arithmetic rather than skill. There is no way to detect it from outside unless the vendor volunteers the count, which is why asking is worth doing.
Where does the broker fit into this?
More than buyers expect. The same strategy on two brokers can produce materially different results because the costs differ, the execution differs and the swap rates differ. An EA sold with results from one broker’s conditions is not portable, and nobody says so.
Which makes the broker choice part of the strategy rather than a separate decision. Research sites such as The Investors Centre, which opens and funds live accounts with its own money to test UK trading platforms rather than compiling rankings from providers’ published fee schedules, measure the costs that determine whether a marginal strategy is viable at all.
That kind of testing establishes what a venue charges and nothing more. It cannot tell you whether a set of rules has an edge, and a cheap broker does not rescue a fitted strategy, so use it to set the hurdle rather than to clear it.
How should you actually evaluate one?
Run it forward, on a demo, for long enough to cover conditions the backtest did not, then on the smallest live size your broker permits. Compare the live results with what the backtest claimed for the same period. Divergence is the finding, and it usually appears within weeks.
Budget for the possibility that it does not work, because the base rate says it will not. Paying a few hundred pounds to find out is a reasonable research cost. Deploying meaningful capital on the strength of a sales page is hope with a download link attached to it.
How much of this is the vendor’s fault, and what should one cost?
Less than the tone of this article implies. A genuine developer with a genuine edge faces the same evidential problem as a charlatan: they cannot prove it without a long live record, and building one takes years during which they earn nothing. The format is a response to a hard constraint, not necessarily to bad faith.
The asymmetry is that the buyer cannot tell the two apart, and the market pays the same either way. That is a lemons problem, and it is why the good developers are hard to find rather than why they do not exist.
Lower than most are sold for, and structured differently. A one-off licence fee aligns the vendor with making the sale and nothing afterwards, which is precisely the wrong incentive when the product’s value is entirely in whether it keeps working. A monthly subscription at least gives the developer a reason to care about month seven.
The structures that align best are the ones almost nobody offers: a share of profits, or a fee that lapses if the strategy underperforms. Their absence from the market is the clearest signal available about how much confidence sits behind the average sales page. Vendors who genuinely believed in the edge would want the upside.
What happens to a working strategy over time?
It decays, usually. Edges in liquid markets come from someone else’s inefficiency, and inefficiencies get arbitraged away as more capital finds them. A strategy that worked for three years and then stopped is the normal life cycle rather than evidence that it never worked at all.
Buying an asset with an unknown remaining life
This matters for how you buy. A strategy is a depreciating asset with an unknown remaining life, and buying one sits closer to leasing equipment than to installing software. Anyone pricing one as a permanent solution has misunderstood what they are purchasing, and any vendor selling it that way is either inexperienced or counting on you not noticing.