|
PureBytes Links
Trading Reference Links
|
This posting provides some food for thought. If one could devise a
function which accurately simulates the random and non-random aspects of
markets using convenient data measures (price, volume, volatility) then it
would be possible to train a neural net using other than historical data.
Simulated data would provide a much more extensive training data set thus
improving the NN's predictive power and avoiding curve-fitting on limited
data.
I don't believe that there is any point training a neural net, or any
predictive system on random data since by definition, there is no
relationship between the training data and future data.
But then I don't believe that markets are entirely random.
-- John
__________________________________________________________
John T. Nelson | John's Trading Journal
trader@xxxxxxxxxxxxxxx | http://trader.computation.org/
__________________________________________________________
Hosted by Lexonia Internet Services
http://www.lexonia.net/
|