441 MQL5 Gold EAs Backtested Under Identical Conditions: The Statistics

441 MQL5 Gold EAs Backtested Under Identical Conditions: The Statistics

"How well does a typical gold EA actually perform?" No single sales page can answer that. So we backtested the XAUUSD (gold) Expert Advisors sold and distributed on the MQL5 Market under one shared set of conditions, and counted the results.

This page covers the 441 EAs that completed the test. We do not name individual products; we publish only the overall distribution.

Test conditions

Every EA was run under the same conditions.

We tried 469 EAs under these conditions and 441 ran to completion. The other 28 were excluded: 22 timed out and 6 stopped because they flooded the log. When an EA was measured on several timeframes, we kept one row per EA, preferring the run on its design timeframe.

The overall picture

The median EA made about +$2,650 over 11 years with a profit factor of 1.21 and a 14% maximum drawdown. But the spread is enormous, so the idea of an "average EA" means very little.

Profit and loss breakdown

Six in ten finished in profit, yet roughly one in four nearly wiped out the account. Blow-ups are not rare exceptions.

Maximum drawdown distribution

The distribution piles up at both ends. EAs either keep drawdown small or sink past 80%, and the middle is comparatively thin. Grid and martingale-style systems, which look shallow most of the time but fall hard once they break, make up much of the second group.

Profit factor and trade count

66 EAs placed fewer than 100 trades in 11 years. With that few trades, chance dominates, and the result says little either way.

Applying a simple screen

Keeping only EAs that are profitable, have a profit factor of 1.3 or more, a maximum drawdown under 30% and at least 100 trades leaves 178 EAs (40%). More than half fail even this lenient screen.

Beware results that look too good

21 EAs showed a maximum drawdown under 2% while growing net profit past $100,000 (ten times the deposit): eleven years of almost never losing while multiplying the account many times over.

One fact matters here. Every one of the 441 binaries we tested was built in 2026. That means the entire 2015–2026 test period is the past that each EA could already "see". A strategy fitted to known history can be made to look as smooth as you like, so the more extreme the numbers, the more you should suspect that fitting.

We describe screening by build date and live track record in We Real-Tick Tested 800+ Market EAs — How Many Survived?.

Limits of these statistics

• These are 1 minute OHLC results. The model is coarser than real ticks, so scalpers in particular can look better than they would in reality.
• Default inputs were used. They may differ from the set files or account conditions the authors recommend.
• This is a historical test. As noted above, every EA was measured on a period before its build date, so these numbers do not predict future results.
• The 28 EAs that did not complete are excluded, so the real failure rate may be higher than these tables show.

How to use these numbers

When you evaluate an EA, check where its advertised results fall in this distribution. A result like "1% maximum drawdown and 100x growth" is an outlier in the top few percent of 441 EAs, and it is also the typical shape of curve fitting. A profit factor around 1.2 with a 10–20% maximum drawdown, by contrast, is perfectly ordinary.

Always read performance figures together with maximum drawdown. Looking at net profit alone makes an EA that nearly blew up look like a winner. Our testing approach is published in our real-tick verification report.

If you cite the figures on this page, please link to this page as the source.