Neural Network sites - any good?

Discussion in 'Automated Trading' started by shogunmike, Nov 19, 2007.

  1. Hi guys,

    I was wondering if anybody has any opinions on sites that are promoting Neural Network type forecasting methods such as:

    http://www.stock-forecasting.com/
    http://www.stockneuromaster.com/
    http://www.tradecision.com/

    I'm curious as to whether NNs work at all in trading. Having read some of the posts here, I get the impression that people believe they're not a substitute for "finding an edge" etc, but they are a useful tool.

    Any thoughts?

    Thanks,

    Mike.
     
  2. ronblack

    ronblack

    Stock-forecasting.com has a free predictor. It predicts for DIA OCLH for today the following:

    132.00 131.89 131.48 132.18 +0.36%

    Let's see. Too good to be true. Easy to evaluate this software but I'm afraid it is pure linear regression or curve fitting.

    I prefer data mining which is also sort of a NN method. The only one I know is APS:

    www.tradingpatterns.com

    They have a demo and you can see what signasl/patterns you get for a number of days.

    I plan to try tradecision soon. It looks quite interesting.

    Ron
     
  3. I liked the look of TradingPatterns.com, the data mining approach looks good.

    So you think stock-forecasting.com is one to miss? I'm just wondering where to spend my hard earned cash. It looks like a good deal for only $40 a month. Tradecision is a lot more expensive!

    Thanks for your comments,

    Mike.
     
  4. ronblack

    ronblack

    No, I didn't say that. I will evaluate a few predictions first. I may subscribe. Thanks for the link.

    However, I noticed something curious. They calculate the accuracy of their prediction based on the stock price value vs predicted value and not based on the predicted change from previous close vs the actual change.

    For example, if a stock closed at 156, up 6 and they predicted 151, up 1 then according to their method the accuracy is: 96.8%. But based on the absolute change from the last close the accuracy is 16.6%.

    So if they insist that they forecast price then they may claim what appears to be high accuracy but that means little to traders, it is not high enough.

    I will have to evaluate several predictions to see how they perform.


    Ron
     
  5. Ron,

    I agree that it is a strange way to calculate accuracy...

    ...strange from a trading point of view, but not from a marketing point of view! ;-)

    However, they do provide previous prediction accuracies. I will also try it out for a while (maybe paper trading with their method), and see how it goes.

    Mike.
     
  6. ronblack

    ronblack

    I agree. I will try the demo for a few days before I subscribe. I may came up with a way of increasing the accuracy or using their results to construct my own indicators. Probably contrarian:)

    Ron
     
  7. LT701

    LT701

    maybe there is something to the best Neural networks stuff, but there's no question in my mind that it's a favorite topic of con men
     
  8. gwb-trading

    gwb-trading

    Mike,

    If you are really interested in getting an understanding of Neural Networking and how it can be applied to the market, then you may want to consider getting the NeuroLab add-on for Wealth-ab 4.0 and taking it for a spin.

    http://www.wealth-lab.com/cgi-bin/WealthLab.DLL/getpage?page=NeuroLab.htm

    Doing this will provide you with a solid educational background about the nut & bolts of Neural Networking.... and you can determine yourself if NN provides any edge for your trading.

    Best Regards,

    - Greg




     
  9. Digs

    Digs

    Forget NN.

    There is not a NN anywhere that can pick a turning point in the future.

    A price series is near impossible to pick a turning point from historical data and maths.
     
  10. "Near impossible" or "impossible"? If "impossible", then this wouldn't this conclusion apply to any attempt to pick a turning point, not just neural networks?
     
    #10     Dec 23, 2007