my AI knows the future and the future belongs to AI :)

[michal.kreslik]

My dear kreslik.com ladies and gentlemen,

I haven't been posting much lately, however, I've received numerous inquiries regarding the status of my current trading endeavors.

I presently operate a private investment firm where 100% of the trading activities are executed through a fully automated process, leveraging internally developed artificial intelligence models. These AI models have been meticulously trained on an extensive dataset, facilitated by a robust machine learning cluster, featuring a multitude of GPU servers, SQL servers and a lot of other infrastructure.

A friend asked me today about some specific patterns on EUR/USD, prompting me to share an illustrative example from today, October 19, 2023. Below, you will find the EUR/USD chart for today, showcasing the AI-generated production trading decisions. Green dots denote buys, red dots denote sells. x axis: time (CET), y axis: price (EUR/USD).

While for obvious reasons I can't go into great detail about the whole AI setup, I'll be happy to answer some of your questions as long as it does not compromise my competitive advantage

Michal

EURUSD_20231019.png (288.6 KiB) Viewed 5045 times

[FrazMan]

Hi Michal, I guess I'll kick the ball off with a couple questions.

1) Could you please provide a couple topics which you recommend ppl read/familiarise themselves with in order to contribute to a fruitful discussion?

2) Which type of NN is most applicable to identify patterns in time series? I'm assuming RNN?

3) If you were starting from scratch, what would be a high level process map you would follow to implement an AI strategy I.e. starting with data acquisition -> setting it loose on the market.

Thanks very much for your time.

[BambinoFlex]

My dear kreslik.com ladies and gentlemen,

I haven't been posting much lately, however, I've received numerous inquiries regarding the status of my current trading endeavors.

I presently operate a private investment firm where 100% of the trading activities are executed through a fully automated process, leveraging internally developed artificial intelligence models. These AI models have been meticulously trained on an extensive dataset, facilitated by a robust machine learning cluster, featuring a multitude of GPU servers, SQL servers and a lot of other infrastructure.

A friend asked me today about some specific patterns on EUR/USD, prompting me to share an illustrative example from today, October 19, 2023. Below, you will find the EUR/USD chart for today, showcasing the AI-generated production trading decisions. Green dots denote buys, red dots denote sells. x axis: time (CET), y axis: price (EUR/USD).

While for obvious reasons I can't go into great detail about the whole AI setup, I'll be happy to answer some of your questions as long as it does not compromise my competitive advantage

Michal

EURUSD_20231019.png

How does it manage a losing trade?

[kiwiarian]

My dear kreslik.com ladies and gentlemen,

I haven't been posting much lately, however, I've received numerous inquiries regarding the status of my current trading endeavors.

I presently operate a private investment firm where 100% of the trading activities are executed through a fully automated process, leveraging internally developed artificial intelligence models. These AI models have been meticulously trained on an extensive dataset, facilitated by a robust machine learning cluster, featuring a multitude of GPU servers, SQL servers and a lot of other infrastructure.

A friend asked me today about some specific patterns on EUR/USD, prompting me to share an illustrative example from today, October 19, 2023. Below, you will find the EUR/USD chart for today, showcasing the AI-generated production trading decisions. Green dots denote buys, red dots denote sells. x axis: time (CET), y axis: price (EUR/USD).

While for obvious reasons I can't go into great detail about the whole AI setup, I'll be happy to answer some of your questions as long as it does not compromise my competitive advantage

Michal

EURUSD_20231019.png

Hi Michal,
Starting from scratch, how does this process begin?
Does one need an idea first, for example take Tro's buzzard method where it takes so many consecutive candles in a row before a reversal is due. Is this then the AI model? Do we then ask it to get Stats on say EURUSD and Asian session with this method and determine what's the appropriate number of consecutive candles needed for the reversal trades? Im trying to expand on Tro's Stats indicators here which is the limitation of my knowledge.
I'M guessing we are not talking asking a cloud service such as chatGPT here either, I go round in circles asking it questions and it tells me to go away and code up your own analysis software.
To be honest, cloud AI is probably not the best way to go as your data might end up training the public data model.
So private AI is the way to go.
Sorry for the basic questions, chatgPT has been my exposure to this technology.

[bb01100100]

Thanks for starting the thread, Michal.

Without giving anything away:

• What proportion of your datasets are not financial timeseries? Or more generally, how much of your operation looks "outside" of the market it's trading?
  
• Can you share anything on your approach to training the models? E.g. are you frequently (or even continuously) retraining as sufficient new inputs arise, or does this occur at a slower/fixed frequency?
  
• Do you adjust inputs during operation (sort of a follow-on from above, but I've seen an options market maker update their model's parameters during the course of trading session in order to express a changing view of risk)?
  
• As an operator of such a setup, do you spend time attempting to understand how the models generate their signals, or do you focus instead of validating the fitness of the results? I'm not an ML engineer, but I've read about models becoming difficult for a human to reason about as they become more complex.

Thanks,
bakedbeans

[michal.kreslik]

Ok, I've realized that it might be challenging for me to answer everything without divulging too much of the know-how, but let me attempt to address some of the points you raised one by one.

> Starting from scratch, how does this process begin?

It should come as no surprise that the initial stage in any quant-based process involves acquiring and cleansing high-quality source data. Without proper and clean data, all subsequent steps are compromised.