Neural Networks Revisited

Discussion in 'Automated Trading' started by maninmoon, May 24, 2016.

  1. userque

    userque

    Sweet.

    So you're telling them what you want...and they figure out how to code it?
     
    #131     Jun 3, 2016
  2. well not to be combatant, but different systems' constructions produce different degrees of longevity. i.e. if something is geometrically based then I have a lot more faith in its ability to withstand the dreaded market changes that so many people carry on about.

    but if we're talking an RSI & MACD bout of nonsense then it'd need to be > 500 trades for my own comfort.

    also would depend on avg win V avg loss & % profitable. honestly over 100 trades, those are the only stats that I would personally care about. if I have 65% wins and 1.5x greater win than loss, then I would be fine with accepting it as a good system.
     
    #132     Jun 3, 2016
  3. more or less. this has been going for awhile. I built a framework over the last two years to encapsulate price and then saw many patterns that I needed to be tested. I've got a long way to go with it - check your PMs in a few, I'm almost done writing you there lol.
     
    #133     Jun 3, 2016
    userque likes this.
  4. userque

    userque

    Lol...no worries...I'm battle hardened :)

    Ok, but what about if you knew nothing about the system other than it performed as the chart I displayed early on. How long would those type of returns have to continue before you'd be willing to believe that the system works, or wasn't 'fool's gold,' 100 years, 1 year, Never, something else?
     
    #134     Jun 3, 2016
  5. For comfort, I think it should work on out-of-sample data of 2 years. AND it should work across a set of instruments (say 5+) that have been selected PRIOR to testing the models on them. Selection bias in the instruments that are chosen for modelling is a very easy/common mistake.
     
    #135     Jun 3, 2016
  6. userque

    userque

    Yes, I agree. That would be comfortable.

    But there is no comfort for the question at hand. :)
     
    #136     Jun 3, 2016
  7. OK I finally reread the image.

    you're near the point of statistical significance with 27 trades. your losses are very well limited. give me my sample size of 30 and i'd be good!
     
    #137     Jun 3, 2016
  8. Getting back on topic. Has anyone managed to build a simple NN that gives a reasonable trading performance across a set of instruments?
     
    #138     Jun 3, 2016
  9. userque

    userque

    #139     Jun 3, 2016
  10. #140     Jun 3, 2016
    userque likes this.