I would like to introduce you to my web application

Discussion in 'Trading Software' started by gpsignals, Mar 15, 2017.

  1. gpsignals

    gpsignals

    Hi everyone. I just released the beta of Evolutionary Signals. This web site allows you to build stock market prediction models using AI (genetic programming to be specific).

    The main idea of the site is to make stock market model building as accessible as possible, without the need to be a programmer. The primary user defined parameter is determining what to use as predictors (Gold price, interest rates, etc.), and the AI does most of the rest.

    I am also including a crowdsourcing component which involves the ability to buy and sell subscriptions to models on the site, and to improve the AI based on the overall results achieved by the user base

    The models are currently limited to buy/sell predictions in a small number of indexes. I am looking to expand the types of models and predictor series drastically over the coming months. The UI needs work and I have a large feature set I didn't include yet, but I would welcome any input or feedback.

    I am in the process of trying to build a small initial user base and push this towards a GA release, and, most importantly, determine the level of interest in such a tool.

    The URL of the site is https://www.gpsignals.com


    Thanks for sitting through this pitch. I would greatly welcome any opinion or feedback.

    David Moskowitz
     
  2. kandlekid

    kandlekid

    Well done, and very interesting. Reminiscent of the Collective 2 site. Just curious, what do you get out of
    it, if subscribers are paying those who created models using your software ?
     
  3. Great job!
     
  4. themickey

    themickey

    Can you run us through this how a member benefits using the following analogy.

    Prospector Pete is in the desert prospecting for gold (trading) when his you beaut metal detector he personally built comes upon a large pile of gold nuggets.
    Other than digging these up himself and becoming rich, how can he generate more profits by going onto someone's web site and 'building a gold nugget prediction models using AI'?
     
  5. jharmon

    jharmon

    Which data provider do you use? Signals are garbage with garbage data providers. Your blog talks about Quandl - I sure hope you don't their Yahoo data (which comes from CSI).
     
    marketsurfer likes this.

  6. Interesting idea, David. I do have some questions. What measures have been taken to assure the accuracy of the data feeds? I have worked with folks in this space who use "watson" type supercomputers to run the genetic evo algos. What is your computer power?

    Thanks
     
  7. gpsignals

    gpsignals

    Thanks all for the positive comments. This is my first exposure to this forum, but it looks like a great place for knowledge exchange in a productive manner.

    KandleKid,

    The current model is based on monthly subscriptions with allow a certain number of models and subscriptions. Model creators will receive payment based on the number of subscriptions they receive. Originally, I toyed with the idea of payments per subscription (at a cost set by the creator) and a split between site and creator. That turned out too complicated and too much of an initial hurdle. Right now, my intention is to prove the concept before dealing with payments.

    Themickey,

    I would be renting him the prospecting tools. Perhaps he does not know how to build or maintain these tools himself.

    Jharmon,

    Right now, I am using Quandl for a handful of popular indexes and indicators, mainly because it is free and within my current budget. I come from a market data background, so I am aware of these issues and the available data sets. I don’t believe the current sets I am using and the granularity level I am looking at is highly successful to minor data errors (I am not yet incorporating individual stocks or calculating breadth.)

    Marketsurfer,

    My current algorithm doesn’t currently have an issue with computer power, as I don’t run millions of individuals for millions of generations (as that would likely produce overfitting in the data I am looking at).
     
  8. tommcginnis

    tommcginnis

    I regret I had to breeze through it way too fast this morning, but I couldn't figure out *what* was supposed to supply inputs (dependent variables) to any model...

    Any help? Yes, I saw the "Model Parameters" area, but do not recall seeing any sort of *selection*.


    As far as UI, yeah, I saw some bumps, but so what -- the overall architecture looks --
    -- sound
    -- obvious (and therefore, *learn-able*)
    -- robust (which, to distinguish it from "sound" might be taken to mean, "scalable")

    Still, and back to my big question: how to designate inputs?!?

    I'll give it another shot after close......
     
  9. gpsignals

    gpsignals

    tommcginnis,
    Thanks for the feedback.
    You select the predictors (independent variables) on the model parameter screen as shown below. It will default to S&P500 as the predictor (as well as the investment target). This is probably the clunkiest part of the interface but the one that it is hardest to get perfect.


    upload_2017-3-16_15-36-59.png
     
  10. tommcginnis

    tommcginnis

    So, to "predict" the S&P500, I choose from 1y, 10y, 2yr,.... etc down that list?
    So, I'm choosing the path {a,b,c,...,..i} by matching-or-lagging {a,b,c,...,..i}??

    I'll take another look right now..... but it sounds like a lot of trouble to compute an MLE of a minimized-differences surface across {potential} estimators.

    AH!
    http://digitalcommons.morris.umn.edu/cgi/viewcontent.cgi?article=1001&context=cs_facpubs
    Gotcha.

    Neat. But not something I need right now.
    But if you're interested in this sort of thing, I came across this guy back in 1985
    https://en.wikipedia.org/wiki/Nicolas_Rashevsky
    and ran a small library of his stuff through the ol' copy machine.
    VERY neat -- and if you can find him being cited, you'll find the papers that cite him are one you'll want to have read.
     
    #10     Mar 16, 2017
    Xela likes this.