Is Neural Network Useful to Analyse Stock Market?
I'm busy on cleanup the Neural Network code after World Cup. Now the performance is acceptable, momentum and rprop learner for MLP were done too. RBF Network almost works.
The Neural Network is not magic, I just use it to do non-linear regression. The results that I got are totally based on my unserstanding to stock market other than neural network.
Anyway, as the primary motivation for BlogTrader project, the neural network and other machine-learning tools (such as SVM) will be continually under developing. I hope to release next version of BlogTrader Platform in later September. Maybe a sample neural network indicator will be included in this coming version. If you are interesting in these approachs, you can write your own indicator based on NN.
Again, I try my understanding on stock market for 000001.ss, here's a result just for joy:
The yellow line is output of neural network in training period, the BLUE line is the prediction for un-trained period, the prediction uses none high-low-open-close values of predicion period, it's a mid-term prediction:
Click on the picture to enlarge it
Posted at 04:15AM Jul 24, 2006 by dcaoyuan in Trading | Comments[6]
NNs are not useful from what I understand. They are simple a black box non-linear regression. The common problem is over-fitting/curve-fitting. Robustness is essential. Most Traders use discretionary TA oder rule-based automated trading systems. Therefore a backtesting functionality is most important (next to charting functions). It's said that genetic/evolutionary algorithms are useful in combination with neural nets or trading rules. Investoxxx is a software which combines all of these but it's expensive. A cheap software for backtesting is Amibroker. However it lacks AI capabilities. Anyway the main difficulty is always parametrization. Which work best, which are stable? Each trader has to judge using his own market experience. Btw, you can look up a guy named Adrian Parusel who uses EViews for linear regressions to select the inputs and Investoxx for neural net predictions. It seems to work for German Bund futures but those trend so clearly that I wonder what he needs NNs for. What literature do you read for machine learning programming? I am interested in that stuff, too.
Posted by abc on July 26, 2006 at 05:55 AM PDT #
Hi,
For literature on machine learning programming, no other ones are needed, the literatures on SVM are enough:)
Posted by dcaoyuan on July 30, 2006 at 10:24 PM PDT #
For what value is this the prediction ?
Posted by ulrich on September 13, 2006 at 06:01 AM PDT #
I have downloaded this file but cannot get it to open properly. Do I need another application?
Mike
Posted by mike on May 27, 2007 at 05:22 AM PDT #
I fount a Neural Network very useful, is just necesary to apply the correct training model.
We must Remember Neural Networks are just a brain emulation.
Regards.
Posted by D.R on October 28, 2007 at 08:43 AM PST #
good
Posted by 221.220.187.44 on December 15, 2007 at 12:43 PM PST #