Results 1 to 6 of 6
  1. #1
    Rookie Member NeuralTip's Avatar

    Talking Hey guys, I'm PT !!!

    Hows it going, I'm PT (nickname)

    Joined Communitraders because I've been a trader for a while (new to binary options though). I specialize in Machine learning and AI which I use extensively in my trading strategies.
    Saying hello, wondering if there are any other techies.

    Thanks,
    PT

  2. #2
    Moderator Kolyo's Avatar
    Quote Originally Posted by NeuralTip View Post
    Hows it going, I'm PT (nickname)

    Joined Communitraders because I've been a trader for a while (new to binary options though). I specialize in Machine learning and AI which I use extensively in my trading strategies.
    Saying hello, wondering if there are any other techies.

    Thanks,
    PT
    Hi NeuralTip! Welcome to CommuniTraders!


    You are actually the first trader here, specialized in AI. We will be very interested to know how you think the last achievements in your field of interest could be applied to binary options trading. I know that there are aims to use neural networks and genetic algorithms for predicting the market movements. What do you think about that ideas?

  3. #3
    Specialist Member marvel's Avatar
    Quote Originally Posted by NeuralTip View Post
    Hows it going, I'm PT (nickname)

    Joined Communitraders because I've been a trader for a while (new to binary options though). I specialize in Machine learning and AI which I use extensively in my trading strategies.
    Saying hello, wondering if there are any other techies.

    Thanks,
    PT

    Hi NeuralTip!

    We are really interested in you opinion about the possible application of AI techniques in BO trading. Are you planning to work in the financial industry after you finish the your machine learning degree? I am very interested in strategy backtesting but still don’t have very good programming skills to do it properly. I will appreciate very much if you share some links to useful resources from such articles and web sites, where it is explained in simple and understandable way the basics of this extremely powerful field.

    Regards,
    Marv

  4. #4
    Rookie Member NeuralTip's Avatar
    Quote Originally Posted by marvel View Post
    Hi NeuralTip!

    We are really interested in you opinion about the possible application of AI techniques in BO trading. Are you planning to work in the financial industry after you finish the your machine learning degree? I am very interested in strategy backtesting but still don’t have very good programming skills to do it properly. I will appreciate very much if you share some links to useful resources from such articles and web sites, where it is explained in simple and understandable way the basics of this extremely powerful field.

    Regards,
    Marv
    I do not have a ML degree, I am an engineer turned software architect. I work for a medical software company doing medical diagnostic and medical billing AI.

    I think machine learning is underutilized by a lot of traders in the market, and no I don't consider some of the neural network implementations included in software like NinjaTrader count really. Their implementations are really basic and years behind the state-of-the-art for current algos.

    A good example of an implementation with some ML algos is one I wrote a article for on my trade tips site/analysis blog. I posted it in the strategies section but here is the direct link: http://neuraltip.com/turning-news-in...ated-approach/

    Another place recurrent neural networks mentioned in the link above are useful is deep learning of chart patterns. See most trading software featuring neural networks use them to directly predict next day prices/close or direction/momentum using range normalized values for prices and volume instead of variance normalized. Even worse, some use the raw price without any normalizing or data preprocessing.

    Think about how you as a trader consume price and volume data and derived technicals. Do you linearly follow each open, low, high and close price of your time frame from the beginning trying to predict the next day or week? Or do you use different time frames to try and determine market sentiment, trend and momentum according to scope? For example, you probably check the weekly bars of your stock/derivative and the weeklies for an index to get a feel of what longer term sentiment is in play. You probably check dailies with a range of 30 days or more to get an idea of the trend in play and maybe 15 minute bars of the last week to get an idea of the momentum in the present and past.

    Most importantly when you look at charts, your mind doesn't pay attention to the exact differences in every bar period's price/technicals, you abstract a general concept about the market and the instrument you are trading based on the patterns/tendencies you see compared to your experiences in the past. If you think about it, this is evident as the truth because there is no succesful trading software that doesn't show you a visual chart/representation of the price/vol. and technicals. Why doesn't anyone buy software that just gives you a spreadsheet view of the data? Because your brain doesn't care about the noise, you are extracting abstract patterns.

    You can have the neural network you use do the same thing, by A. making it a recurrent deep neural network that abstracts the data patterns in layers and B. Making sure your inputs are normalized and presented to the algorithm that allows it to abstract patterns over a picture or window of the time series data. This way the patterns that the neural networks learns and bases it's outputs of buy or sell or what have you is not "if price before t-1 < t-2, t-3 > t -7..." but "Double Bottom Pattern in 15 min/3 day window, Teacup Pattern in 24 hour/30 days window...". This generalizes much better than the former and allows you to multiply the effectiveness of your data because now you can learn from any stock/market traded instrument as long as it has some indication of price/volume etc.

    When you talk about back-testing, I see it as a mistake that most people unfamiliar with ML make. In ML, the gold standard of performance is always for data that you didn't train on because whatever strategy you created, you created by deriving information from the same data you are back-testing from!. This means that your strategy is a statistical superstar in back-testing but if any new type of ground truth cause and effect in the market is introduced (and this happens on a daily basis these days), it will immediately rocket your error rates. That's why the real measure of a strategy is it's ability to generalize, as a rule, if your strategy cannot be applied to more than a few stocks or even a one stock/derivative, chances are it's over-fitted and is lying to you on how good it is.

    This is why I recommended earlier to do recurrent NNs that actually learn the patterns in the data in an abstract fashion, even better (only if you ignoring news in your model) is to make sure you only train your network between news releases of sufficient time apart so that you can be confident that the market has incorporated the news into it's price and the price fluctuations only represent emotions (greed/fear), which are much much easier to model with NNs or anything else really than trying to model the ground truth cause and effect that drives the market.

    I'm tired of typing, but if you like these insights, check out my blog and sign up to NeuralTip, I plan to write stuff like this I think about on NeuralTip's blog.

    Thanks,
    PT
    Last edited by NeuralTip; 11-20-2013 at 07:30 PM.

  5. #5
    Rookie Member NeuralTip's Avatar
    Also, forgot to say, if you guys want access to tips I get from my neural network implementations you can sign up to my newsletter, beta is on time to launch 12/01.

    Thanks,
    PT

  6. #6
    Junior Member allanrock's Avatar
    Thank you Neural Tip for the good explanation of the neural network application in the financial markets and binary trading as well. I would like to switch from simple discretionary trading to more advanced algorithmic trading, but my knowledge is very small yet and I am not so convinced that I will be able to move to that new and exciting area so quickly. I will appreciate very much your help and your guidance, especially with some literature and readings where I should start.

    All the best,
    Allen

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •  
3