Are AI based EA better than purely algorithmic / TA based EA?

New comment
 

Hi!

I know many already use AI to help them place orders manually through chats and report easy gains. 
On here i see many EA in the topCharts are AI based promising to interact with AI to place and manage trades for you.
Since there are usually no backtests possible i would like to start this thread for actual users of these AI based EA to tell us if they are profitable and/or more profitable than purely algorithmic/TA based EA.

Just in practice i think the response times from these online AI like chatGPT, deepseek, anthropic, google are quite high. 
So asking them on a tick/bar basis frequency will become unfeasable in practice. 
Can they beat neural networks on using TA trained models ?

Rgs

Andreas

Forex Market – App-Store von Handelsrobotern für MetaTrader 5, Expert Advisors und technische Indikatoren
Forex Market – App-Store von Handelsrobotern für MetaTrader 5, Expert Advisors und technische Indikatoren
  • www.mql5.com
Offizielles Schaufenster von Anwendungen für den Handel im Terminal
 

If some user told that his EA is using AI (or based on AI) so the first thing to do is to look at this user: is he well-known user on this forum? does anybody here know him and believe in him (trust him) to be ready to invest some fund to him in such as "AI EA"? or not?

Because it is easy to tell anything without any single proof and without beeing responsible for that for example.

-------------------

And the discussion about Market products is prohibited on the forum.

------------------

It was some months ago that the service desk made the cleaning the Market from "AI EAs" - they removed those products together with the sellers/users.
I think - they (the service desk) will do/repeat it from time to time.

------------------

As to AI so I found one post -

Forum on trading, automated trading systems and testing trading strategies

Sma 14 based system (5 minutes)

William Roeder, 2025.03.14 18:50

ChatGPT (the worst), “Bots Builder”, “EA builder”, “EA Builder Pro”, EATree, “Etasoft forex generator”, “Forex Strategy Builder”, ForexEAdvisor (aka. ForexEAdvisor STRATEGY BUILDER, and Online Forex Expert Advisor Generator), ForexRobotAcademy.com, forexsb, “FX EA Builder”, fxDreema, Forex Generator, FxPro, “LP-MOBI”, Molanis, “Octa-FX Meta Editor”, Strategy Builder FX, “Strategy Quant”, “Visual Trader Studio”, “MQL5 Wizard”, etc., are all the same. You will get something quick, but then you will spend a much longer time trying to get it right, than if you learned the language up front, and then just wrote it.

Since you haven't learned MQL4/5, therefor there is no common language for us to communicate. If we tell you what you need, you can't code it. If we give you the code, you don't know how to integrate it into yours.

We are willing to HELP you when you post your attempt (using Code button) and state the nature of your problem, but we are not going to debug your hundreds of lines of code. You are essentially going to be on your own.

ChatGPT
  1. Even it says do not use it for coding. * 
  2. Mixing MT4 and MT5 code together.
  3. Creating multiple OnCalculate/OnTick functions.
  4. OnCalculate returning a double.
  5. Filling buffers with zero in OnInit (they have no size yet). Setting buffer elements to zero but not setting Empty Value to correspond.
  6. Calling undefined functions.
  7. Calling MT4 functions in MT5 code.
  8. Sometimes, not using strict (MT4 code).
  9. Code that will not compile.
  10. Creating code outside of functions. * 
  11. Creating incomplete code. * 
  12. Initialization of Global variables with non-constants. * 
  13. Assigning a MT5 handle to a double or missing the buffer and bar indexes in a MT4 call. * 
  14. Useing MT4 Trade Functions without first selecting an order. * 
  15. Uses NULL in OrderSend (MT4). * 
bot builder Creating two OnInit() functions. * 
EA builder
  1. Counting up while closing multiple orders.
  2. Not useing time in new bar detection.
  3. Not adjusting for 4/5 digit brokers, TP/SL and slippage. * 
  4. Not checking return codes.
EATree Uses objects on chart to save values — not persistent storage (files or GV+Flush.) No recovery (crash/power failure.)
ForexEAdvisor
  1. Non-updateing global variables.
  2. Compilation errors.
  3. Not checking return codes.
  4. Not reporting errors.
FX EA Builder
  1. Not checking return codes.
  2. Loosing open tickets on terminal restart. No recovery (crash/power failure.)
  3. Not adjusting stops for the spread. * 
  4. Using OrdersTotal directly.
  5. Using the old event handlers.

 
Andreas Alois Aigner:

Hi!

I know many already use AI to help them place orders manually through chats and report easy gains. 
On here i see many EA in the topCharts are AI based promising to interact with AI to place and manage trades for you.
Since there are usually no backtests possible i would like to start this thread for actual users of these AI based EA to tell us if they are profitable and/or more profitable than purely algorithmic/TA based EA.

Just in practice i think the response times from these online AI like chatGPT, deepseek, anthropic, google are quite high. 
So asking them on a tick/bar basis frequency will become unfeasable in practice. 
Can they beat neural networks on using TA trained models ?

Rgs

Andreas

I don't use such eas but its not impossible . 

In my very unprofessional opinion , if someone tasks an agent with developing a vocabulary , a language , for market formations then they could even do it without being a coder . But , that system would still be prone to overfitting just like our little primitive neural nets, just like a student who has memorized Pythagoras theorem and can only solve the lessons.

But it could be done with reasoning too , again i guess , for instance if that student is in front of a test then he will have to recall the Pythagoras theorem and reason , or , parallelize with the lessons to solve it.

These fields are unique (trading , crypto , betting) in the sense there is a giant multibillion corporation discovering things and they don't need to make money from forex , sort of implying the brains at the tip of the spear don't need to make a living here. 

But then again , those brains did not rely on being "Go" champions for a living so.

 
Lorentzos Roussos #:

I don't use such eas but its not impossible . 

In my very unprofessional opinion , if someone tasks an agent with developing a vocabulary , a language , for market formations then they could even do it without being a coder . But , that system would still be prone to overfitting just like our little primitive neural nets, just like a student who has memorized Pythagoras theorem and can only solve the lessons.

But it could be done with reasoning too , again i guess , for instance if that student is in front of a test then he will have to recall the Pythagoras theorem and reason , or , parallelize with the lessons to solve it.

These fields are unique (trading , crypto , betting) in the sense there is a giant multibillion corporation discovering things and they don't need to make money from forex , sort of implying the brains at the tip of the spear don't need to make a living here. 

But then again , those brains did not rely on being "Go" champions for a living so.

From what i can tell sofar when you are interacting with AI through their API you will most likely give them a price history and a set of technical indicators and then ask it how a professional trader would place orders in that environment (buy/sell). Sofar i havent seen it being able to use a broad macro view of the market or be able to incorporate news or anything like this. I would think that training a neural network or letting an algorithm decide on the same set of inputs would lead to the same result (actually more consistenly probably). 

I have experimented with LLM Time Series models which have been developed in particular for time series (IBM model for example). But from my personal tests i havent seen them be able to predict very accruately. They are looking more like a fitting model to existing data. Sofar i havent seen any working time-series prediction model based on LLM myself.

 
Sergey Golubev #:

If some user told that his EA is using AI (or based on AI) so the first thing to do is to look at this user: is he well-known user on this forum? does anybody here know him and believe in him (trust him) to be ready to invest some fund to him in such as "AI EA"? or not?

Because it is easy to tell anything without any single proof and without beeing responsible for that for example.

-------------------

And the discussion about Market products is prohibited on the forum.

------------------

It was some months ago that the service desk made the cleaning the Market from "AI EAs" - they removed those products together with the sellers/users.
I think - they (the service desk) will do/repeat it from time to time.

------------------

As to AI so I found one post -


Thanks for info. So is it not allowed to build AI EAs? Or what rules are they breaking in order for them to be removed?

 
Andreas Alois Aigner #:

From what i can tell sofar when you are interacting with AI through their API you will most likely give them a price history and a set of technical indicators and then ask it how a professional trader would place orders in that environment (buy/sell). Sofar i havent seen it being able to use a broad macro view of the market or be able to incorporate news or anything like this. I would think that training a neural network or letting an algorithm decide on the same set of inputs would lead to the same result (actually more consistenly probably). 

I have experimented with LLM Time Series models which have been developed in particular for time series (IBM model for example). But from my personal tests i havent seen them be able to predict very accruately. They are looking more like a fitting model to existing data. Sofar i havent seen any working time-series prediction model based on LLM myself.

Maybe its the prompt . If they ask it to trade like a professional trader then the "latent" space it will access are written forex ads , direct or indirect.

If you give a neural net all the data and it has to predict the trading action instead of the next word , isn't it possible that the best indicator could form in its hidden layers?

 
Lorentzos Roussos #:

Maybe its the prompt . If they ask it to trade like a professional trader then the "latent" space it will access are written forex ads , direct or indirect.

If you give a neural net all the data and it has to predict the trading action instead of the next word , isn't it possible that the best indicator could form in its hidden layers?

One inherent issue i see is that LLMs like you say are based purely on words/characters and not on numbers. You have to give the LLMs the method to interpret numbers since they are not inherently related to each other like characters in words. So yes i would think neural networks are at least one methodology that is designed to work with numbers, so i think should outperform AIs (my guess). I would love to see a comparison side by side of the two given the same set of inputs/constraints/trading logic and also compared to a fixed algorithm that follows the same trading logic.

 
Andreas Alois Aigner #:

Thanks for info. So is it not allowed to build AI EAs? Or what rules are they breaking in order for them to be removed?

I do not know ... it may be allowed if we see them on the Market...

I know that the service desk is cleaning the Market from time to time (periodically) from the products having "AI" in the name (I think - the service desk is thinking that name "AI ...EA" is just the marketing only and not the real AI EA for example).

...what rules are they breaking in order for them to be removed?...

I think - they (some sellers) named their products as "AI EA" or similar but in reality - some products had nothing to do with any AI (except this word "AI" in the name of it) ... because I remember - it was some discussion on the forum and complaints from the buyers to the service desk, and it was the post of admin on the forum many months ago about it (about "AI EAs" which do not have any AI at all anywhere ... just "AI name" only ...).

 
will all due respect to all views and opinions, I have 2 years of experience working both algo and AI EAs and I dont think that all AI based EAs are better than algo EAs. 

i have 2 reasons for this: 

1) most if these said "AI Experts" are not well developed, as developing and training an AI model is much more complicated than people would think, and usually all these EAs will lead to underdeveloped and not properly trained models. 
2) for most strategies and methods of trading, algo EAs are perfectly fine in my opinion. in fact, im a firm believer that with a bit of experience in coding and creativity in coding, you can come up with EAs that can do things even better than humans, without the use of AI models. 

would love to hear back your opinions, thanks. 
 
It could all be down to how it's engineered. Imagine making it capable to trade only based on historical data? that's what a fraud would do. The reality is in immediate data, volatility, momentum, order book ticks... I don't see where AI has an edge for finding an entry better than in safe-proven trading systems. But I do see an edge in how an AI can find patterns, price psychology, and how it can manage trades based on changes of news and economic data releases. Is the EA designer a real trader? you have to ask that too... 
New comment