How to backtest with 100% quality

[Manpreet Singh]

Hello experts,

Can anyone suggest how to test an EA with 100% quality, whenever I tried to backtest my terminal didn't show more than 90%.

I don't know is it normal or I am making mistakes.

[Carl Schreiber]

To me this, quality of quotes, is of minor importance for the success of an EA. I would check certain market phases like trend, flat, strong trend reverse, trend => flat, flat => trend and heavy increases of the volatility (like the last days for EUR and USD due to the change of the interest) or big jumps of courses (GBP on 26.09.2022)

[Michael Reuben Msidada]

Your question is a little vague, I don't know if you mean quality of testing data or quality of testing in general as mimicking real world conditions. But if you mean quality of testing data, you going to need to download accurate tick data from data vendors. Brokers data has lots of gapes. Check this this article https://www.mql5.com/en/blogs/post/746240 for that.

if it's about trying to match real world condition, I would advise to stop processing all tick and move to process bar closes, with that you can get very close to real world conditions. There is a lot of online materials talking about the subject. 

[Manpreet Singh]

Thank you both of you

[CyberBot]

Greetings friends, from us in Indonesia.

We had the same experience as you asked about. It's frustrating. As a result of determining the backtest time limit; other than that about the very extreme difference in results if we use MODE "Control points..." with "Every tick..." which is more accurate but takes a long time.

The solution we managed to find is:

A. The backtest still uses the “Control point…” mode, it's just that the program is set to only be active & loop once an hour [adjusts the test period used; current H1 timeframe].

B. Displays the results of the algorithm visually, so our eyes can see whether the execution was carried out by EA as expected or not.

C. Don't be satisfied just yet, if the backtest is only 1 month, because we need an EA that is flexible in all conditions. The longer the better.. even up to 2.5 years

[definitely more than enough, so that the algorithmic procedures that we insert have a great chance of happening again in the future]

D. Lastly, assume a 50-50 win over EA work; then use the power of "probability" i.e. activate as many EAs as possible on an account with a small deposit, with time intervals.

Hopefully useful..TQ [Translation from Indonesian]

---

Explanation based on test results: https://www.mql5.com/en/blogs/post/751579   

---