Prediction on Another Stock Index by Neural Network
This time, let's try to apply the neural network on Shanghai Stock Exchange Index(000001.SS on Yahoo! Finance). The network has 2 hidden layers, the 1st one has 5 nodes, the 2nd one has 2 nodes. Output layer has 1 node.
After 5000 epoches training with 2400 sample data, the result is showing below.
It seems the pattern of chart is better than the turning time. Particularly, the backward generalization gets a very interesting result.
When I do the prediction, I do not use the price in the prediction period to adjust the network or as input any more. That is, I only use the price data in the training period to recognize the pattern, then apply the pattern backward or forward directly. So the prediction is fairly a long term prediction.
Click on the picture to enlarge it
Posted at 04:30PM Jun 11, 2006 by dcaoyuan in AIOTrade | Comments[7]
the prediction seems to work terrible outside the training set - as usual for nns. what (still 3?) inputs do you use? and btw, are you planning to release further versions of the source code?
Posted by 134.34.35.28 on June 13, 2006 at 07:17 PM PDT #
I forgot to mention, when I do the prediction, I do not use the price in the prediction period any more. That is, I only use the price data in the training period to recognize the pattern, then apply the pattern backward or forward directly.
Posted by Caoyuan on June 14, 2006 at 01:20 AM PDT #
The 'inverson' is fairly typical when using any predictive method. While inverson's can cause problems they can also be called in advance w/ methods like elliott waze and wilder's delta. It's these periods of disagreement between the methods that the best opportunities are presented.
Mark
Posted by Mark on June 15, 2006 at 03:05 AM PDT #
I have written some code to predict stockquote tendency using FANN library. Now I'm tring to integrate it into Visual Chart software (It's software like HumaiTrader).
More info (spanish) here;
http://open-trader.blogspot.com/2006/07/inteligencia-artificial-un-sistema.html
Posted by Ripe on July 09, 2006 at 11:38 PM PDT #
I agree with one of the comments that neural networks is not the way to go. If you use Backpropagation algorithm and you minimze MSE only you will always overtrain NN or find some
low probability model. There are some ways around this, but still NN is to primitive to find some really cool patterns. And definately it is not more sophisticated than human brain (not this one at least). It would be great if you could continue releasing new code :)
Posted by aa on July 17, 2006 at 03:28 PM PDT #
ABLE
Posted by CZ on October 21, 2007 at 07:30 PM PDT #
I found a little problem refreshing data of the Neural Network i am asume you are using the Backpropagation model, so the software is going to regenerate the neuron set i think.
Regards
Design Engineer
Neural Networks Models and Algorithms
Posted by D.R. on October 28, 2007 at 09:54 PM PDT #