What we did

Update, 3 November 2026: after publishing we found 85 EAs whose converted versions produce exactly the same backtest as another EA on every market — for example 14 unrelated EAs with identical results. The conversion evidently lost what made them different, so their 298 runs are no longer counted, and the figures below have been recalculated. Most of them were profitable on gold, so the gold numbers fell the most. All figures here come from the test on hourly data from May 2025 to May 2026. Since 4 November 2026 every EA is re-run each month on the latest twelve months, so the numbers on the EA pages differ slightly from this article.

The EA Arena catalogs 1,515 Expert Advisors published on the MQL5 CodeBase. For 1,063 of them we translated the MQL5 source into Python automatically and backtested it on one year of hourly data (May 2025 to May 2026) for EUR/USD, GBP/USD, USD/JPY and gold. Every EA got the same account: $10,000, 1:100 leverage, the EA's own lot size (100,000 units per currency lot, 100 ounces per gold lot), one order at a time, the same commission, and a broker-style stop-out when equity falls below 10% of the starting balance.

Most EAs could not be tested

Automatic conversion is hard. Out of 4,252 EA-and-market runs:

  • 2,348 came from EAs whose entry logic the converter could not translate (custom indicators, unusual order handling). We label them "not convertible yet" instead of pretending they were tested.
  • 144 depend on a custom indicator the converter replaced with a placeholder, so they do not test the real EA.
  • 155 converted but never traded. For 112 of them we can point to the conversion bug: the fast and slow settings of a crossover were lost, a price-based indicator was asked to cross zero, lines on different scales were compared, or no entry rule survived.
  • 88 stopped with an error, and 96 traded too little or produced a flat curve.
  • 298 runs from 85 different EAs came out exactly identical to another EA's run on every market, which means the conversion lost what made them different. We no longer count them.
  • 1,123 results are reliable enough to rank.

The results we trust

  • About half were profitable: 562 of 1,123. The median result was roughly zero over the year, with a median maximum drawdown of 7.8%. The typical EA had a longest losing streak of 6 trades and held a position for about a day.
  • On the three currency pairs the typical EA went nowhere: the median result was −0.12%, and only 27 of 830 currency results made more than 10%.
  • Gold looks better than it is. Gold rose 44% over the test year, so long-leaning systems were carried by the trend: the median gold result was +4.5% — with a median drawdown of 60%. 61 of 293 gold runs (21%) lost the whole account.
  • Consistency was rare: of 271 EAs with reliable results on all four markets, 30 were profitable on every one.
  • 187 results had a Sharpe ratio above 1. Over a single year that is easy to reach by chance, which is why the EA Ranking also shows the recovery factor and how many markets an EA made money on.

Categories and risk

The arena's risk review rates 34 of 40 martingale EAs and 14 of 33 grid EAs as "dangerous": they increase exposure after losses or keep opening positions against the trend. In our backtests martingale EAs did not show the usual very high win rates (median 49%), most likely because the automatic conversion trades a fixed lot size instead of reproducing the lot multiplication. Their real-world risk is therefore higher than these numbers suggest — read the risk review on each EA page, not only the backtest.

How to use the EA Arena

  • Treat "not convertible yet" as "unknown", not "bad".
  • Look for EAs that hold up on several markets, not one spectacular result. The "Markets" column in the EA Ranking shows this at a glance.
  • On gold, check the drawdown before the profit: 0.1 lot of gold on a $10,000 account is a much bigger bet than 0.1 lot of EUR/USD.
  • Check the risk rating and the risk points before the P&L.
  • Free code is a learning resource. If you trade an EA, test it yourself in the MT5 Strategy Tester with your broker's data and costs — and you can compare the result with the arena in the Strategy Lab.