Advanced Methods (Neurals Networks, Genetic Algorithms, Fuzzy Logic, etc...)

Discussion in 'Trading Software' started by jaym, Apr 29, 2004.

  1. jaym

    jaym

    Plase post you thoughts on any software you have tried or currently used that include these methods. I would like to hear people's opinions on the various commercial packages that are available and on the methods themselves.

    The three packages I know of are:

    -NeuroShell Trader Pro

    -Trading Solutions

    -Neuo-Lab Add-on for Wealth-Lab

    Anyone used these? Have you used other packages?

    Also what net types do you find most useful in real time trading? How do "evolved" genetic solutions compare to exhaustive search or monte carlo search methods? Do any of you use fuzzy logic or neurofuzzy solutions?

    Please add anything you think would be useful.

    The final questions is advanced methods: Useful in real trading or not?

    ______________
    Jay
     
  2. jaym

    jaym

    If you voted other what software do you use?
     
  3. ramora

    ramora

    Take a look at SNNS at

    http://www-ra.informatik.uni-tuebingen.de/SNNS/

    SNNS runs on Windows or Linux, has excellent documentation, and more features than most commercial products. Although SNNS is not new it is well supported. Many SNNS features are not available in products designed for the trader listed in this thread.

    The commercial packages help in pre-processing data and a user friendly interface. However, anyone who is going to succeed with NN is probably not going to use standard RSI, CCI, Sto, MACD, etc., etc.

    You can use SNNS to learn the basics by developing a simple model in Excel, save the data to text, and then build the NN using SNNS. Try it before spending money on one of the commercial packages.

    Good luck,
     
  4. flakac

    flakac

    Personally, I use Matlab. There is a good neural network toolbox and fuzzy logic toolbox to. The construction of trading model is straightforward, however the problems is connection to data providers and brokers for trade execution
     
  5. These kind of methods came in to vogue some 8-10 years ago with great fanfare, they have not worked, so most don't use them. There are lot of tool boxes available to do this but most are high priced crap.
    So you might be wasting your time chasing such tool boxes or methods.
     
  6. cosmic

    cosmic

    for most of what novices search in this area I would absolutely agree with you.

    But if you choose your goals / targets with NNs carefully you can use them to your advantage, even in demanding iday trading.

    best,
    cosmic
     
  7. Excellent link, ramora!
    Didn't know about this NN tool before, and found another and newer version implemented in Java - JavaNNS - http://www-ra.informatik.uni-tuebingen.de/software/JavaNNS/welcome_e.html .

    JavaNNS provides a GUI on top of calls to a DLL for NNS. This makes it really easy for anyone working in Java to integrate this great tool.
    Using JBuilder with the JADClassFileNode from Softamis - http://www.softamis.com/ie/ot/jad/download.html you can easily decompile and view class-compiled jar-packages.
     
  8. i use the neural network located b/w my ears -this one seems best suited (tho more often than not the logic is a bit fuzzy :p )
     
  9. jaym

    jaym

    What specifically do you use them for? Are you using the nets for price prediction or classification of patterns?

    I am a novice to this arena but I would like to test these methods out and see if I can create some viable trading strategies. What I don't want is to get caught buying over priced commercial software that turns out to be limited or too black box in nature.

    Any suggestions?
     
  10. pspr

    pspr

    In the past I have wasted a lot of money and time trying to get something reliable out of a few of the better known neural net softwares. I eventually moved away from them after becoming frustrated with the time commitment and lack of consistent results.

    When the developers of the software can't even show you a system they have developed that performs one loses interest in being able to do any better.

    Also, those who have made NNets work for them have to retrain the net regularly and most nets fail in short order making you start the whole process over.

    Maybe others have different experiences with NNets. If so please chime in.
     
    #10     May 7, 2004