On Apr 15, 11:53=A0am, "news.kornet.net" <acegen...@[EMAIL PROTECTED]
> wrote:
> I am currently developing a simple time series analysis program to test
a
> small set of data. I want to test and compare results of Moving Average,
> ARMA, and ARIMA. Is there anyone who has Java source for ARIMA? Because
I =
am
> using Java, I prefer Java. But, FORTRAN or C/C++ sources would be fine
too=
..
>
> Thanks
Good Direction ..
Make sure that the models you build/estimate
1 are invertible
2. have a residual mean that is zero everywhere ( i.e. no pulses,
level ****fts , seasonal pulses or local trends in the residuals )
3. the residuals have constant variance i.e. no structural break
points or dependency on the level of the observed series and can't be
represented by an ARIMA process
4. the parameters don't change over time ... simply break the series
down to mutually exclusive subgroups and test the hypothesis of
constant parameters ala Greory Chow
5. don't incluse any non-significant parameters
6. the final acf of the residuals ( after you have verified 1-5) is
free of structure.
Finally compare your models with FREEFORE a freeware package available
from http://www.autobox.com/freef.exe
Dave Reilly
Automatic Forecasting Systems
http://www.autobox.com
215-675-0652 in the US
|
