Machine Learning – AI – Backtest Module

• 01/01/2022 at 7:03 AM #184225
  Khaled

  Hello everyone,

  Happy new year 2022!

  I read the various threads on Machine Learning and especially the code developed by juanj https://www.prorealcode.com/topic/machine-learning-in-proorder/ or the use of polynomial equations as presented by Degardin Arnaud https://www.prorealcode.com/topic/xbtusd-regression-1-machine-learning-obtain-decision-trees/ which requires exporting data to a 3rd party software and generate the various factors for the equation, not simple and requires regular updates of the factors.

  I’ve tested the code of JuanJ and the results are improved when I apply it to one variable. However, I’m not sure that, on the long run, it finds the optimal set of variables when I apply it to two (or more) co-dependent variables, i.e. if the behavior of Indicator1 based on [Period1] (e.g. to take or exit Long) influences the behavior of Indicator2 based on [Period2] (e.g. to take or exit Short), then I’m not sure if the optimum is reached under this co-dependency.

  May be one of the avenues to explore with PRT in 2022 is that they develop a piece of code that continuously simulates the Backtest, even with limited number of variables (say max 4) and limited time span (say 1 month look back), so that we can include this code into the systems and optimize these 4 variables in real-time up to the previous candle. I think this will give an edge to PRT on the market, at least for individual traders. The backtest module is already there, so we just need to access this possibility through ProOrder in real-time. I’m not sure if this is already possible with the Arrays. If so, please share an exemple on how to continuously optimize two (or more) variables in real-time.

  I’m not sure what’s the best way to convey this suggestion to PRT. If this idea is supported, may be Nicolas or Roberto can reach out to PRT?

  

  Khaled

  Participant

  Topics: 59

  Replies: 314

  Been thanked: 65 times
  01/01/2022 at 8:28 AM #184228
  GraHal

  Great idea … anybody can send Suggestions to PRT, they invite us to do just this using the Form on the link below …

  https://www.prorealtime.com/en/contact?suggestion=1

   

  1 user thanked author for this post.

  Khaled

  

  GraHal

  Participant

  Topics: 204

  Replies: 8441

  Been thanked: 2049 times
  01/01/2022 at 8:43 AM #184231
  Khaled

  Suggestion sent to PRT. I’ll share the feedback when they replay. We keep chasing the variables, be it MA Periods, Breakeven, SL, etc. I think this can be a game changer for all of us.

  1 user thanked author for this post.

  GraHal

  

  Khaled

  Participant

  Topics: 59

  Replies: 314

  Been thanked: 65 times
  01/01/2022 at 1:06 PM #184241
  Nicolas

  Khaled wrote:

  The backtest module is already there, so we just need to access this possibility through ProOrder in real-time.

  It means huge memory consumption on servers and very difficult way to identify why an order has triggered or not because of the variables values optimized at that time. BTW, I have never met something similar on any other platform.

  I think it could somehow be soft coded in a strategy. Of course, more variables would mean more complexity.

  1 user thanked author for this post.

  Khaled

  

  Nicolas

  Keymaster

  Topics: 81

  Replies: 20830

  Been thanked: 4744 times
  01/01/2022 at 2:10 PM #184245
  Khaled

  Happy New Year Nicolas!

  Glad to hear your opinion. I agree, it’s a tricky topic… I think the blocking point will be the Intellectual Property. But PRT can protect this module like ROCnRoll indicator for example.

  I’m sure I’m not the only one who spends his days backtesting. I found myself becoming more of a programmer than a trader (to be). The market structure changes frequently. Even the most robust and most sophisticated strategies that I found on this forum do not perform over time, either losing money or in the best case doing way below Buy & Hold. So updating the parameters, at least SL – Exit strategy parameters, is a must. There will be always exceptions. I’ve a small system which generated >500% in 2021 in Demo without any optimization during the year. It’s a pure chance that I found the idea on internet and I was extremely lucky to find the best exit strategy that works with NQ. Tested on DJ, SP and DAX, it doesn’t provide good results.

  “very difficult way to identify why an order has triggered or not because of the variables values optimized at that time”. I’d say it’s under the client responsability and discretion to use the module or not. Clients can sign a waiver. If you use ROCnRoll in a system, you don’t know what’s the code and as far as I know PRT doesn’t take responsability for losing trades.

  You never saw something similar on any other plateforme… This is why I’m saying that it will give a serious edge to PRT. From marketing standpoint, it will make a serious difference and brings more clients. If this requires that they invest in more processing capacity, why not? There’s a cost/benefit analysis that PRT should perform.

  I’d very much like to know how big Trading houses address this optimization issue. If anyone has a clue, please share with us.

  

  Khaled

  Participant

  Topics: 59

  Replies: 314

  Been thanked: 65 times
• Author
  Posts