Has anybody tried Neural Networks? Does it work?

Discussion in 'Strategy Development' started by mizhael, Apr 1, 2010.

  1. Any successful stories?

    I am starting to play around with Neural networks.

    Please point me to good papers/articles/softwares about how NN can be used and pitfalls can be avoided...

    I am also interested in seeing how NN can be coupled with entry/exit and position sizing...

    Your pointers are greatly appreciated.

    Thank you!
  2. No, it is a waste of time. But there are internet discussions giong back over 15 years. And by the time you are done, you will find it will not add value.
  3. Evidence pls. Thx
  4. I disagree. It's no more a waste of time than any other tool. The key to successfully applying NN techniques (or any other) is to first find an approach that has a statistical edge already. NNs are a potentially convenient way to tell when conditions line up to give an edge. They're also adaptable if you use back-propagation; as conditions change, the parameters can be made to self adjust.

    It's no different than applying DSP techniques like FFTs or Butterworth filters to financial analysis. People get all excited and think it's a panacea and others say it's completely useless. Again, there are many tools and techniques available, but none are panaceas nor are they useless. It all depends on your background and what you're interested in.
  5. I followed a guy for a while who claimed to be an expert on NN. His picks weren't good.
  6. byteme


    Sounds like conclusive proof.
  7. LOL!

    I like neural nets, just don't expect to throw in a price series and get a proper forecast.
  8. eot uses one. It needs some work, but I think the idea is solid. Anything attempted to predict is going will usually be wrong on unexpected news. For example, the original news break on Greece.
  9. That's what I have been told and seen. Garbage in garbage out.
  10. Exactly what I said before. You have to put some bounds on what you're measuring.

    For instance there have been statistical studies of candlestick patterns. It would be straightforward to train a NN to find which 1-, 2-, 3- etc candlestick patterns are predictive. For instance certain candlestick patterns have a 60% or 70% reliability for daily bars for stocks. A NN could find if there are candlestick patterns with similar reliability on other instruments and time frames.
    #10     Apr 1, 2010