Input/Output Data Convertion in BackProp.

[Antonio J. G. Abelem [Marco]]

	I'm using neural networks to predict financial time series, 
specifically the gold-prices' time series. 

I have experienced some problems with the convertion scheme used to 
present data to the network.

I have mainly used LINEAR CONVERTION (original data value converted to 
the ranges 0, 1 or -1,+ 1), but it does not seem to work properly.

	I have also had some attempts with: 

a) input data in its original value
b) in its derivative form
c) the input minus data average
d) the percent difference between input Ti+1 and Ti

However, for all these cases the target patterns need to be converted 
before presenting to the network once its  output is between 0 and 1 
(to the sigmoid) or between 1 and -1 (to the hyperbolic tangent).

My results for the single-step mode are good, but i think it could be 
better.

On the other hand, for the multi-step, the results are very bad.


Any suggestions on that will be very appreciated.


Thanks.


Antonio