Is AI-assisted EA development actually improving Gold (XAUUSD) strategies or just making overfitting easier? - page 2

New comment
 
Muhammad Minhas Qamar #:
If you're using AI, you have access to vast amount of knowledge that can genuinely help you come up with better strategies, instead of letting it wander in its own direction. Might aswell learn about quantitative finance, why are markets believed to be stochastic processes, why prices are believed to be random. Answers to such questions will deepen the understanding and approach towards strategy-creation. 
Good point, thanks for sharing your perspective. There is definitely a lot to learn on the quantitative side of things appreciate you adding to the discussion!
 
Daniel Andrew Clay #:
 

The Monte Carlo testing step is something I have not be using. I do backtesting and some forward testing but I haven't been systematic enough about stres stesting the randomness factor. That's clearly a gap I need to close.

The point about forward testing on a small live account for a longer period being the most critical step really resonates. I think a lot of people including myself early on treat a demo forward test as equivalent and it just isn't. The execution, the spread behavior, the slippage it all changes the picture.

Exactly, that’s the gap.

Monte Carlo really shows how fragile a strategy is, and live testing proves what actually holds up. Demo just isn’t the same, especially with XAUUSD. I totally agree on that point with you

 

I think that first, you need to define what you actually want to trade. Otherwise, determining whether the AI-based EA bot is truly working would be very difficult.

You need to understand what you are looking for in order to know where to search.

 
IMHO, the more you deviate from a "raw" Single backtest, the further you deviate from the underlying concept that inspired you to create the EA in the first place. I always try my hardest to follow the KISS principle and Murphy's Law. When in doubt, I clear the slate and start over from scratch.
 
In my opinion, there are 2 possibilities for using AI in software, either EA or other applications.
1. AI is only used to help create concepts or create program code.
2. AI becomes part of the System

If it is only used as a tool, of course it will make it quite easier to create applications and at this level the programming knowledge required by humans can be said to be quite standard, but if AI is to be included in the application then advanced programming knowledge is needed. 
For example, if AI is only used as a tool to create an EA, then it is enough to create it using the functions in the MQL 5 editor, but if you use AI in the program then there needs to be an API or Python etc. which is not enough just the functions in the MQL 5 editor.

Even though it is said that AI makes programming easier, in reality AI often makes mistakes, both syntax errors and logic errors. 
In my opinion, up to now, we can't let go of AI and just create an EA from start to finish and we're running straight away.

After the EA is ready, of course it will be backtested and like the process of creating other systems, revise both input variables and trading techniques or filters as needed.

What you must always remember is that AI can make mistakes.

In my opinion, there is no guarantee that software or EAs created by 100% human thought will be successful during live trading and we know that backtest results do not guarantee future results.

In my experience, when doing a backtest, comparing the backtest results with live ones, we often see that when we lose, we immediately assume that the EA is bad, the design is wrong, etc., even though if we look at the backtest results, especially from the graph, we will see that the EA is sometimes profitable and sometimes it is a loss. From a loss to a profit again, sometimes it takes time, which can be fast or slow. During the backtest, we don't have any emotions there, we see the graph just going up and down, even going down sharply and then rising slowly, but during live when there is a loss, emotions arise because of defeat, Equity decreases and the EA is considered wrong, even though there is a possibility that in the future the EA will give a profit like the results in the Backtest. 
It's up to us as humans to be patient or not.
 
Daniel Andrew Clay #:
Good point, thanks for sharing your perspective. There is definitely a lot to learn on the quantitative side of things appreciate you adding to the discussion!
Yes
 
Using an AI to program an EA is like having a senior developer on your team; the code will be practically flawless, and the protocols (opening, closing, trailing, signals, etc.) will be academically applied and that is precisely where the hitch lies.

There are several types of people in this community:

  • coders who are only good at coding (and who do it masterfully, those with a lot of MQL rating points on their profile)
  • traders who are only good at trading (and who do it masterfully, those with the DO's actually)
  • coders learning to trade
  • and traders learning to code

The truth is, if you don't know the market, you're going nowhere, and unfortunately, that can only be learned through direct experience; there is no book, no wikipedia article, no tutorial for it. And when facing the market, AIs are just as helpless as the best of the aforementioned coders.

Besides, what I'm saying is so true that there is a project that set out to hand over the reins to AIs for trading (the link was deleted by moderator) where Claude, Gemini, ChatGPT, DeepSeek, Groq and Qwen face off... and the results aren't exactly stellar.


 
Daniel Andrew Clay #:
 

The Monte Carlo testing step is something I have not be using. I do backtesting and some forward testing but I haven't been systematic enough about stres stesting the randomness factor. That's clearly a gap I need to close.

The point about forward testing on a small live account for a longer period being the most critical step really resonates. I think a lot of people including myself early on treat a demo forward test as equivalent and it just isn't. The execution, the spread behavior, the slippage it all changes the picture.

monte carlo is essential but be aware it is still testing past data


imo if its logical then fine usually the less filters the better in my experience

 
I believe AI is changing the world completely by aiding new strategies I love how I use AI in my trading and the best part is mql5 also allow the use of ai in mql5 language 

 
Icham Aidibe #:

There are several types of people in this community:

  • coders who are only good at coding (and who do it masterfully, those with a lot of MQL rating points on their profile)
  • traders who are only good at trading (and who do it masterfully, those with the DO's actually)
  • coders learning to trade
  • and traders learning to code

Don't forget to list prompters who are only good at prompting!😂

I know, I know, kids... "vibe coding."😎

123
New comment