Has anyone had success coding an EA with AI application? | Page 2 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

Does anyone has script to run success rate of fib pivot points 0 replies

I Will Learn Coding in 1 Month - Give me the 80/20 of MT4 Coding 14 replies

Has anyone had luck with high frequency using either neural nets or genetic Pr? 3 replies

Backtesting Discussions - I had a 99.72% success on a strategy. 18 replies

Has anyone had problems like this with MB trading 0 replies

Attachments: Has anyone had success coding an EA with AI application?
Exit Attachments

Has anyone had success coding an EA with AI application?

  • #21
  • Sep 6, 2025 4:36am Sep 6, 2025 4:36am
Quoting 919gilead
Testing update, and done for the day {image} NB Please I am not selling anything and neither I am soliciting for anything. I am just showing Clive and anyone trying to use AI to code their EA that it is possible.
Hi Dude

Looks like you've got this cracked, amazing results.

I am a bit shocked at the size of the code for sure but it appears to work so what the heck!!

Great to see it can be achieved using AI and you're saying the same as I've found you have to keep it under control , do a discussion session first before each section and save the Mq5 files separately in date order so when the occasional cock up comes I just go back one save file save and restart him from there.

I think I'm gonna persevere for the time being with Chatgpt5 as I feel we are getting there be it fairly slowly.

All the best Clive
Feel the Fear and do it anyway!!!
  • #22
  • Sep 6, 2025 5:13am Sep 6, 2025 5:13am
I will calculate on your 25 pips exit a day. That sounds interesting. Still - if you need coding power, just let me know.[/quote]

Thanks for the offer I might very well take you up on it if ChatgpT5 drives me to the edge!!! Do you code MQ5 by the way or MT4 doesn't matter eithers fine , but the one I'm working on is in MT5 as I thought it was more versatile and faster for scalping or account copying as I'm sure I will need to split accounts several times to manage trade size and getting orders filled efficiently.

Anyway let me save you some time with the calculations around making 25 pips per session as I did ask it to work out if I started with £1,000 and traded at max size and made just 25 pips per session how long would it take to make a £1,000,000.

Answer 31 weeks!!! So in theory in just over 6 months I'd be a Millionaire!!! Please see attached calculations.

So all I have to do now is make an EA that makes 25 pips per session based on London and New York opens.

Can't be that hard to write an EA make 25 pips can it??

Keep in touch.
Attached File(s)
File Type: csv £1,000 balance weeks to million.csv   < 1 KB | 181 downloads
Feel the Fear and do it anyway!!!
  • #23
  • Sep 6, 2025 12:57pm Sep 6, 2025 12:57pm
Quoting Iamfx
Interesting. That's quite a massive code as far as I'm aware although I did read about one recently that was 20k lines. If possible I would think about removing you pics as people run scraper codes and the like and could copy aspects of your system.
Thanks Iamfx. I think and trade outside the box, so if anyone can copy any aspect of my system from the screenshot or discussion, be my guest.
Low Risk Low Stress High Reward High Probability Trades
  • #24
  • Sep 6, 2025 4:57pm Sep 6, 2025 4:57pm
Quoting sahitya666
I started using ChatGPT for EA creation back when it was ChatGPT-4, and I must say — AI has improved a lot. I always create my EAs with ChatGPT, not because I don’t know coding, but because it speeds up the process. One thing I’ve noticed with AI is that when coding, you need to clearly explain the logic. Instead of just saying ‘create an MA crossover,’ I specify it like ‘if the long MA is above the short MA.’ That way, it generates fewer errors, sometimes none at all.
Just curious, do you have this option?

Seems it would / could / should make a big difference in the quality of the coding or any provided information.

Attached Image (click to enlarge)
Click to Enlarge Name: 2025-09-07_7-54-47.jpg Size: 17 KB

It takes a lot for me to IGNORE you, but you can try.
  • #25
  • Sep 6, 2025 8:27pm Sep 6, 2025 8:27pm
Quoting clivepackham
I will calculate on your 25 pips exit a day. That sounds interesting. Still - if you need coding power, just let me know.
Thanks for the offer I might very well take you up on it if ChatgpT5 drives me to the edge!!! Do you code MQ5 by the way or MT4 doesn't matter eithers fine , but the one I'm working on is in MT5 as I thought it was more versatile and faster for scalping or account copying as I'm sure I will need to split accounts several times to manage trade size and getting orders filled efficiently. Anyway let me save you some time with the calculations around making 25 pips per session as I did ask it to work out if I started with £1,000 and traded at max size and made just 25 pips per session how long would it take to make a £1,000,000. Answer 31 weeks!!! So in theory in just over 6 months I'd be a Millionaire!!! Please see attached calculations. So all I have to do now is make an EA that makes 25 pips per session based on London and New York opens. Can't be that hard to write an EA make 25 pips can it?? Keep in touch. {file}[/quote]

Don't over optimize, don't over leverage.
"Wisdom is proved righteous by its works"
  • #26
  • Sep 7, 2025 2:06pm Sep 7, 2025 2:06pm
HAce tiempo que no escribo, en FF, no soy codificador, pero yo he estado usando Gemini para codificar, aunque he tenido tropezones he logrado terminar mis sistema. Era un sustema complejo y al principio tenia problemas por el codigo extenso, el plan que tengo es el PRO, asi que decidi crear librerias .mqh. Para trabajar. Si quieres tener un sistema complejo talvez te sirva esta idea, separar las funciones en librerias. Asi las puedes usar en cualquer indicador o EA que crees. Lo mejo de todo es que cualquier estrategia puedes tenerla en una libreria, quieres una mezcla de varias estartegias? No hay problema solo pon tus reglas y el EA se convierte en el cerebro que dirige todo. Quieres usar el mismo EA en varios activos? Lo haces multitiframe y multisimbolo. Asi monitoreas el mercado desde un solo sitio.

Quieres crear un control de noticias para que el EA no opere o cualquier accion, no hay problema, pones la logica dentro de una libreria y ya no teienes que volver a codificar esta funcion en ninguo de tus nuevos EA.

Quieres operar solo en tendencia o en rango? Crea una libreria con tus indicadores favoritos, y listo.
Quieres un sistema complejo de manejo de ordenes, SL, TP, TS, etc, grid, martingala? Crea librerias.

Asi obtendras un sistema multifuncional y portable.

Que sea rentable? Todo depende de la logica que apliques, La IA no te hara gananr dinero en automatico, lo que lo hara es un sistema bien pensado y bien configurado.

Saludos amigos
IH - B221 Return Today: -0.2%
  • #27
  • Sep 7, 2025 5:55pm Sep 7, 2025 5:55pm
I love AI's, find that if one doesn't do proper job, just take project to next one. Using 3 AI's at present, coding and me don't go together, so this option came so handy.
Runing indicator and ea at present, both created by AI's , will see how it goes. I find when building something, starting from base and than building step by step, if I ask for every thing initially, there is mess.
Attachment are result's from last Week(one day), today starting with trend, analasys done by one AI.
Attached Image (click to enlarge)
Click to Enlarge Name: Performance, no trend.PNG Size: 44 KB
  • #28
  • Edited 9:05am Sep 9, 2025 8:53am | Edited 9:05am
this is just update from that testing, since GPBJPY had best result's, so I've setup 4 strategies, 5min chart, this am I have analasys done by AI(just to mention SL=2 or 3 pips didn't work well. latency on this one is 30ms, gona move it to EC2 server now, 6ms to see if it makes difference):
GBPJPY Strategy Performance Analysis - Updated Stop Losses
Strategy Definitions

  1. Lot 0.05: Countertrend, Dynamic SL = 1.0
  2. Lot 0.11: Trend-based, Dynamic SL = 1.0
  3. Lot 0.13: No trend, Dynamic SL = 0.5
  4. Lot 0.21: No trend, Dynamic SL = 1.0

Performance Results
Lot 0.05 (Countertrend, SL=1.0)
P&L Values: 0.07, -2.18, 1.33, 0.44, 1.53, 0.48, 0.58, 0.04, 0.07, 0.92, 1.19, 1.16, 0.03, -2.29, 0.07, 0.92, 1.36, 0.58, 0.41, 0.37, 0.03, 0.41, 0.14, -2.24, 0.21, 0.24, 0.13
Total P&L: +4.88 Win Rate: 24/27 trades (88.9%) Largest Loss: -2.29 Average Win: +0.60 Average Loss: -2.24
Lot 0.11 (Trend-based, SL=1.0)
P&L Values: 0.07, 0.30, 0.35, 0.30, 1.05, 3.15, 1.73, 0.23, 1.50, 0.75, 0.90, 0.53, 0.22, 0.26, 0.83, -4.73, 0.22, 0.07, 0.44, 0.30, 1.06, 0.38, -4.65, 0.30, 1.19, 0.75, 2.24
Total P&L: +12.42 Win Rate: 25/27 trades (92.6%) Largest Loss: -4.73 Average Win: +0.83 Average Loss: -4.69
Lot 0.13 (No trend, SL=0.5)
P&L Values: -6.02, 0.09, 0.35, 0.35, 3.45, 1.15, 0.45, -5.84, 1.50, 1.25, 1.06, 1.33, 0.97, 0.09, 0.27, 0.89, 1.33, 0.80, -5.66, 0.09, 0.36, 0.27, 0.44, -5.58, 1.50, 2.65, 0.35
Total P&L: -2.52 Win Rate: 23/27 trades (85.2%) Largest Loss: -6.02 Average Win: +0.95 Average Loss: -5.78
Lot 0.21 (No trend, SL=1.0)
P&L Values: -9.71, 0.14, 0.57, 0.57, 5.58, 1.86, 5.43, 3.36, 2.84, 0.17, 0.71, 2.44, 1.51, 0.58, 2.15, 1.29, 0.72, -9.02, 0.43, 5.00, 3.10, 1.29, 0.14, 0.57, 4.00, -9.44, -8.86
Total P&L: +15.81 Win Rate: 23/27 trades (85.2%) Largest Loss: -9.71 Average Win: +2.23 Average Loss: -9.26
Strategy Ranking (Best to Worst)
1st Place: Lot 0.21 (No trend, SL=1.0) - BEST

  1. Total P&L: +15.81
  2. Win Rate: 85.2%
  3. Key Strengths: Highest total profit, good win rate, largest average wins
  4. Risk: Largest losses when stopped out (-9.71, -9.44, -8.86)

2nd Place: Lot 0.11 (Trend-based, SL=1.0)

  1. Total P&L: +12.42
  2. Win Rate: 92.6% (highest)
  3. Key Strengths: Most consistent wins, highest win rate
  4. Risk: Moderate losses when wrong

3rd Place: Lot 0.05 (Countertrend, SL=1.0)

  1. Total P&L: +4.88
  2. Win Rate: 88.9%
  3. Key Strengths: Consistent small wins, controlled losses
  4. Performance: Solid but modest returns

4th Place: Lot 0.13 (No trend, SL=0.5) - WORST

  1. Total P&L: -2.52
  2. Win Rate: 85.2%
  3. Key Weakness: Tightest stop loss (0.5) creates frequent losses that outweigh wins
  4. Issue: Stop too tight for market volatility

Key Observations
Impact of Tighter Stop Losses

  1. 0.5 SL (Lot 0.13): Too tight - creates losses that negate profitable trades
  2. 1.0 SL: Sweet spot for all other strategies - allows breathing room while controlling risk
  3. Countertrend strategy improved dramatically from previous test (was worst, now 3rd)
  4. Trend-based strategy became highly profitable (92.6% win rate)

Risk-Reward Analysis

  1. Best Risk/Reward: Lot 0.21 - larger wins justify occasional larger losses
  2. Most Consistent: Lot 0.11 - highest win rate with reasonable profits
  3. Over-optimized: Lot 0.13 - stop too tight for effective trading

  • #29
  • Sep 19, 2025 1:54pm Sep 19, 2025 1:54pm
Quoting clivepackham
{quote} Hi Dude Looks like you've got this cracked, amazing results. I am a bit shocked at the size of the code for sure but it appears to work so what the heck!! Great to see it can be achieved using AI and you're saying the same as I've found you have to keep it under control , do a discussion session first before each section and save the Mq5 files separately in date order so when the occasional cock up comes I just go back one save file save and restart him from there. I think I'm gonna persevere for the time being with Chatgpt5 as I feel we...
Hey Clive, I trust you're done or almost finished with your coding. I'm done and just tweaking and perfecting my settings.
Attached Image (click to enlarge)
Click to Enlarge Name: Global Prime Demo Report7d - 9192025.gif Size: 166 KB
Low Risk Low Stress High Reward High Probability Trades
  • #30
  • Nov 17, 2025 3:34am Nov 17, 2025 3:34am
Quoting sahitya666
I started using ChatGPT for EA creation back when it was ChatGPT-4, and I must say — AI has improved a lot. I always create my EAs with ChatGPT, not because I don’t know coding, but because it speeds up the process. One thing I’ve noticed with AI is that when coding, you need to clearly explain the logic. Instead of just saying ‘create an MA crossover,’ I specify it like ‘if the long MA is above the short MA.’ That way, it generates fewer errors, sometimes none at all.
My experience:

1. First, you need to understand the logic of the system, step by step.

2. Break down what the system does into steps – one by one.

3. Describe each step; creating an indicator helps, as we need to have control over how everything works.

4. At the coding stage: Wriite "impersonate great constructor of indicators and systems and write a prompt for creating high-level code. First, a prompt to write code, not code.

5. At each stage, we add that "if you have any questions, ask them" — it helps to clarify all doubts and greatly accelerates success. It may seem unnecessary, but it is critically important.

6. You can use several AIs at the same time, for example, one for code and another for reading and explaining what is in the code (sometimes LLM can suddenly change the code during a longer session, and you need to have control over this).

7. After each session, especially unsuccessful ones, we write "Analyse the session in terms of what can be done better next time to achieve the goal faster".

I have written many indicators and scanners, and EA is an interesting challenge that I am now going to tackle.

If you have any questions, maybe I can answer them.
  • #31
  • Nov 17, 2025 6:58am Nov 17, 2025 6:58am
Quoting clivepackham
So in theory in just over 6 months I'd be a Millionaire!!! Please see attached calculations. So all I have to do now is make an EA that makes 25 pips per session based on London and New York opens. Can't be that hard to write an EA make 25 pips can it?? Keep in touch.
Hi Clive, I read part of your thread (“Trading is as simple as ABCD”) and I saw there are a few systems there — one based on TTIndi and another on Hurst bands. Which one would you like to have coded?

In my opinion, the one based on the slope of the moving averages might be simpler. It’s built on a principle similar to the 919Gilead system (MADdash — Moving Average Distance). Last week I read and analysed both threads. Both are promising, and both face the same issue — whether the move will continue.

But overall, it's not a big problem, because the core principles are very strong: Gilead trades weak currency against strong (so this is not a pair trading - it is one currency against another), and in your case we have pairs with a high daily ADR — which means the pair must make a lot of pips in one direction each day. Both very promising approaches imho.
  • #32
  • Nov 17, 2025 8:18am Nov 17, 2025 8:18am
Quoting Iamfx
I sort of found the longer the conversation the worse the answers especially at high server load times. I would then ask it to make a prompt for a fresh chat then start again. Just at the moment Claude opus 4.1 is absolutely horrendous. A code with 1k lines has 50 errors and it can't seem to fix no matter how many goes it has. This is from supposedly the best ai coder. I had to drop chunks of it into chat 5 to fix

Every LLM has a very unpleasant tendency: during a long conversation it can rewrite the entire code from scratch, ignoring earlier orders.

To solve this, try first to refine the algorithm the code should be based on, and then write a very good prompt for the EA. After that, with every new instruction, add that it must adhere to that prompt.
  • #33
  • Edited 10:10am Nov 17, 2025 8:44am | Edited 10:10am
Quoting clivepackham
I will
I’ve just taken a closer look at the system based on the bands.

There’s a problem with it — the red arrow points to a place where the price is rising but the middle moving average is falling (blue arrow).
The moving average should still be rising with the price, which means the indicator is repainting. In my opinion, it’s not suitable for use as part of an EA.

Attached Image (click to enlarge)
Click to Enlarge Name: 2025-11-17_143823.jpg Size: 220 KB
  • #34
  • Nov 17, 2025 1:47pm Nov 17, 2025 1:47pm
Short answer YES, BUT how long it took to get that fully working code is another thing...

Sometimes AI nails it first pop, other times I go round and round in circles for days.. Its like arguing with my Ex wife, just keeps giving same answer, even though I KNOW they are 100% wrong

Ive learnt to start with a simple base code, and keep building from there.. Save each copy I am happy with.. There is a reason things have V42..

I write all my instruction down in a word doc first, with a clear, concise list of variables, inputs colors etc etc Then paste into AI.

If you dont have any knowledge in coding it will be 10000 times harder, I have very limited, but enough to pick up its mistakes like when its using mt4 code, and not mt5...

BUT if you just type, Write me an EA that will turn $10 into 1 mill in two weeks so I can retire, you are going to be very disappointed..
Im only here for the Guzzoline...
  • #35
  • Nov 21, 2025 12:04pm Nov 21, 2025 12:04pm
Quoting 919gilead
{quote} Thanks Iamfx. I think and trade outside the box, so if anyone can copy any aspect of my system from the screenshot or discussion, be my guest.
I think that copying or re-engineering the system without knowing it inside - out is not possible at all, but discussing what it contains could be interesting. I have read almost the entire thread (Maddash) and will try to build a robot based on it. Here, I will summarise the main principles in order to see if anything is missing.
  • #36
  • Nov 21, 2025 12:31pm Nov 21, 2025 12:31pm
Quoting Ettore
{quote} I think that copying or re-engineering the system without knowing it inside - out is not possible at all, but discussing what it contains could be interesting. I have read almost the entire thread (Maddash) and will try to build a robot based on it. Here, I will summarise the main principles in order to see if anything is missing.
A few thoughts that I will add to later.

The strength of this system lies in the fact that we try to play by choosing the strongest currencies against the weakest ones. What does that mean?

Let's say that over the last hour, the yen has been heavily bought on all markets (i.e. there is high demand for the yen). The dollar is being heavily sold everywhere - all dollar pairs are being heavily sold. Something has probably happened, e.g. traders have got some news that has caused them to buy one currency and sell another.

The USD/JPY pair will be a very good candidate for entry - on one condition that this buying and selling situation continues.

Conclusion number one - movements based on strength - weakness can be predictable and strong. As long as... they have not been exhausted, which means we need to have some kind of limit that tells us that the movement has gone too far.

Predictable and strong - because a lot of traders for some reasons want to sell one and buy another.
  • #37
  • Mar 17, 2026 3:54am Mar 17, 2026 3:54am
Hello colleagues,
has anyone received a complete code or complete execution of a given task from AI? By task I mean automating/writing code for a small system or indicator. I personally have several automated systems and in each of them there is a small gap in the logic invisible at first glance. After which, despite detailed explanations, it is not corrected and what has been done is ruined. Thanks in advance
  • #38
  • Mar 17, 2026 6:02am Mar 17, 2026 6:02am
Attached Image (click to enlarge)
Click to Enlarge Name: screenshot.webp Size: 124 KB

Everything you see in the attachment was generated entirely by Copilot.
  • #39
  • Mar 17, 2026 6:20am Mar 17, 2026 6:20am
Quoting 919gilead
{quote} Hey Clive, Yes I just finished using Anthropic Claude Sonnet 4.0 & 4.1 to build a very complex and versatile 13000+ code lines. I subscribed to the Pro plan and quickly realized I needed a higher subscription with what I was developing, and upgraded to the Max plan. I have zero coding skills and therefore patiently followed every instruction to the letter. If there was any ambiguity I clarified before copying and pasting. I explained clearly what I wanted to code, and where necessary I attached chart illustrations. I always tell Claude not...
Great feedback. Time to take it to the next level to do the dirt work! LIVE TRADING! That would confirm the hype by these AIs. ChatGPT is No.1 for such hype!
Grit; Knowledge; Right Mindset; All you need to overcome. KIS!
  • #40
  • Mar 17, 2026 8:15am Mar 17, 2026 8:15am
Absolutely no problem for ChatGPT to code strategies in Java for Jforex platform.
I assume chatgpt can also transform the code for other platforms, but I have never ask for that task.
0 traders viewing now