Strategy optimization result do not match single run test result - page 6

New comment
 
In my case, this issue only happens when I use custom data. I've noticed significant data gaps (over a year) when I analyze the differences between the log results from two test runs on custom data (no error messages). As mentioned by other users before, this happens in the latest build (5430 in my case). In earlier builds, like 5327 and 5370 I had no issues. Important to mention, I run MT5 under a Wine environment (version 10) under Linux Debian. Haven't had the chance yet to test this in a Windows environment.
Files:
mt5_tester_settings.jpg  41 kb
 

piece of crap software, I can see why prop firms all want you to use mt5


I have the same problem, single run is not the same as optimization.. so everytime I think I got a great optimization actually it's not.. what is the point!.. 

you cannot put the blame on the EA coder.. it's clearly mt5 team of losers

 as a senior dev myself, I can tell you the coders of this software are piece of crap like the software itself

 
djsinae2 #:

piece of crap software, I can see why prop firms all want you to use mt5


I have the same problem, single run is not the same as optimization.. so everytime I think I got a great optimization actually it's not.. what is the point!.. 

you cannot put the blame on the EA coder.. it's clearly mt5 team of losers

 as a senior dev myself, I can tell you the coders of this software are piece of crap like the software itself

There plenty of possibilities to self-defeat yourself by wrong coding in an EA.

A bug on the platform is always possible but we need evidence about it.

You should also watch your language, think what you like but be respectful.

 
It's the type of modelling that leads to inconsistent results: 'every tick' & 'every tick based on real ticks'. The other modelling options work fine.
 
Comments that do not relate to this topic, have been moved to "MetaTrading 5 Agent NOT connecting to the cloud".
 
virtualreal #:
It's the type of modelling that leads to inconsistent results: 'every tick' & 'every tick based on real ticks'. The other modelling options work fine.
Can you prove that somehow ?
 
Alain Verleyen #:
Can you prove that somehow ?
I suspect it happens within the OnTradeTransaction() event function. Next week I will try to reproduce this with a modified EA. 
 

Forum on trading, automated trading systems and testing trading strategies

Different Tester Results for Two Computer Using the Same EA input Parameters

Alain Verleyen, 2025.12.05 01:19

Let me be clear. 

The Strategy Tester is a computing system that takes "inputs" (data, parameters, settings, trading environment and "EA" ex5), and process all of these to provide some "outputs" (backtest trades and stats).

This can fail to produce the same results due to several possibilities :

  • Different inputs somehow. If the data feed is the same as well as the start/end dates (and other tester parameters), there are still several others possibilities to have different inputs (non-exhaustive list).
  1. The swap has no historical data, the swaps used in the EA are the ones in memory at the time of the backtest. It's not entirely clear what they are and how they are updated, I didn't investigate in depth. The best practice to avoid different results (if the swap can influence the results) is the start the platforms at the same time on both computers, and even to check the swaps to confirm.
  2. The strategy tester allow custom settings by symbol, it's very easy to change a setting then forget it, for each trading time or margin used. This need to be checked.
  • Different processing.
  1. The EA can contain some code that give different results on each run. There are a lot of possibilities, I will not detail that here and now. Most of the time the EA is a black box that people don't want to share.
  2. Then the Strategy Tester itself (let's take it as a whole as "MT5", though in practice there are different parts implied). That's the easy "culprit" though in practice it's rarely the real culprit. It could happens, that what a bug is. Though all the others points needs to be checked and confirmed as "proof-safe" before an MT5 bug can be investigated.
  3. The computers. It happens too, but it's even rarer than MT5 bugs. For example a faulty RAM. But no, no the speed of the CPU or of the computer in general (unless proven otherwise). Cosmic rays are also possible but rarely investigated as a cause of computer issue.
We have dealt with a lot of similar topics on this forum, statistically, once the inputs are checked, the best chance for the reason behind the different results is the EA code (including possible custom indicators used).

 
Alain Verleyen #:
Can you prove that somehow ?

So far I haven't been able to reproduce these inconsistent results with a modified EA. I will dig into this further once I have more time to do so.


Ryan L Johnson #:
Thank you for sharing this. I have included this header file into my EA and the log reports "no memory leaks".
 
virtualreal #:

So far I haven't been able to reproduce these inconsistent results with a modified EA. I will dig into this further once I have more time to do so.

Thank you.
1234567
New comment