EA's Recommendation ⑥: Thoughts on evaluation criteria to avoid failure when choosing an EA, explained from a developer's perspective on how to read the numbers
Good morning everyone. This is Renmi.
This is the sixth article. The previous post (⑤) was in February 2025, and now it is May 2026... yes,I’ve been slacking off for more than a year.
It’s not just a little gap; it’s been more than a year due to excessive slacking. Before I knew it, over a year had passed. I’m very sorry (laughs).
Last time, I wrote about how the image of FX in society is too negative. For those who haven’t read it yet, please go here.
https://www.gogojungle.co.jp/finance/navi/articles/83155
Now, this time it’s about “how to choose an EA.”
When you look at GogoJungle and other sites, there are really a lot of EAs lined up, right? Many of you may have struggled with the question, “Which one should I choose…?”
So this time, as an EA developer myself, I’ve summarized the evaluation criteria that I think you absolutely must see.
Note that this time the evaluation criteriaassume a single-position EA (an EA that holds only one position at a time). EAs that use grid trading or martingale have very different risk characteristics, so we’ll discuss those on another occasion.
Do not rely solely on backtest numbers
First and foremost, please allow me to say this.
Do not choose an EA by looking only at backtest figures.
As I wrote in a previous article (Here is an article about overfitting), backtests can be optimized for past data, so the numbers can look anything. Be wary of EAs with “Win rate 95%! PF above 10!”
So what should you look at? I’ll explain in order.
Evaluation criteria to look for
① Profit Factor (PF): 1.2 or higher is sufficient
PF is the value of “total profit ÷ total loss.” If it exceeds 1.0, the overall result is positive.
1.2 or higher: sufficiently functional level
1.5 or higher: good level
2.0 or higher: you should be a little wary in backtests
Personally,I think 1.2 or higher is enough.
To give an analogy with pachinko, even the top setting (setting 6) yields about 110% payout. Converting this to PF is about 1.1. PF 1.2 means an advantage better than pachinko setting 6. If that advantage can be reproduced in real trading, it’s a solid weapon.
An EA that maintains PF above 1.5 in real trading for a long period is quite excellent. However, if backtests show PF5 or PF8, beware because that is likely overfitting as I mentioned earlier.
② Maximum Drawdown: in real operation, aim for 20% or less
Drawdown is the percentage drop from the peak value of assets.
To be honest, there isn’t a single correct % lower bound. Maximum DD depends heavily on how much margin you started the EA with. Comparing the DD of a low-capital EA with one running full lots is meaningless.
Neverthelessin real operation, aiming for 20% or less is prudent. If it goes above 20%, it becomes psychologically quite tough, and once it passes 30%, many people think “I’ll stop.” Even EAs with large backtest DD can be workable in real trading if you adjust capital and lot size appropriately. You don’t have to dismiss an EA just because backtest DD is 50%; you can reduce the lot size and operate it.
③ Number of trades: 100 or more per year
This is a point that is easy to overlook.
Suppose there is an EA with a 10-year backtest period but only 20 trades. Even if the win rate is 90%, the sample size is too small for statistical reliability.
Having 100 trades or more per year makes the statistics reasonably reliable.
④ Always verify real-trade performance
This is the most important part.
Backtest numbers are just calculations based on past data. In actual real trading, factors such as spread variation, slippage, and order rejection cannot be reproduced in backtests.
Myfxbook,Realtrade,FX Blue— EA developers that publish real trading history on services like these increase credibility. Be sure to check real performance for at least six months before considering a purchase.
By the way, the EA I sell has real performance data on Realtrade. For example, “GTX” has real-trade data since the end of 2021. I publish it deliberately so you can judge by real numbers as well as backtests.
GTX real-trade performance https://real-trade.tech/accounts/52392
GTX sales page https://www.gogojungle.co.jp/systemtrade/fx/34527
Renmi’s EA list https://www.gogojungle.co.jp/users/189446/products
⑤ Backtest period: ideally over 10 years
Backtests under 5 years may just reflect a period that happened to fit.
EA that has endured various market environments like the Lehman Brothers crisis (2008), Abenomics (2013–), and the COVID-19 crash (2020) tend to be more versatile.
⑥ Look at the shape of the equity curve
Not only the numbers, but the shape of the equity curve (graph) is also important.
✅ Good example: gently rising to the right
❌ Bad example: spikes up and down suddenly, or stays flat for a long time and then jumps
The latter often indicates a pattern of earning mainly during a short period by chance.
⑦ Check the developer’s information and track record
Finally, also look at the developer’s credibility.
Whether they have been selling multiple EAs for a long time
Whether they publish real-trade results
Whether they respond properly to inquiries
Anonymous appearances promising “50% monthly profits!” should raise suspicions.
Summary
To summarize the points for choosing an EA,
PF 1.2 or higher is enough; 1.5 or more in real trading is excellent
Maximum drawdown: aim for 20% or less in real operation (note that this varies with capital size)
Number of trades: 100 times per year or more for statistical reliability
Always verify proven real-trade results
Backtest period: ideally over 10 years
Choose equity curves with a gentle upward trend
Also check the developer’s credibility
Remember: the more an EA’s numbers look too good, the more suspicious it tends to be. I’ve even been able to intentionally create an EA with “win rate 96%, PF 8.57”... (for details,an article about overfittingis here).
Well then, until next time!
Renmi’s EA List
https://www.gogojungle.co.jp/users/189446/products