Sharad,

The approach that you are taking is often insufficient to form an
appropriate model.

For a few reasons ...

1. The model may be different than a p,p-1 model which is the assumed model
when using their automatic aic selection.
2. The residuals from the model may have a non-constant mean requiring
Pulses, Level Shifts , Seasonal Pulses and/or Local Time Trends. Failure to
remove or model these components often (always !) yields a FALSE
AUTOCORRELATION signal
3. The model may have changed over time
4. The parameter may have changed over time ..

etc...

To solve your problem simply download from DOWNLOAD.COM the latest release
of FreeFore
or go directly to the site http://www.autobox.com/freef.exe

For more on automatic modeling see http://www.autobox.com/pdfs/catchword.pdf

If I can help simply send me the series with all of it's characteristics in
an excel file and I will
process it through AUTOBOX/FreeFore

Regards

Dave Reilly
AFS
215-675-0652

-----Original Message-----
From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]
Behalf Of sharad gupta
Sent: Friday, February 13, 2004 1:46 AM
To: [EMAIL PROTECTED]
Subject: [edstat] Time series and correlated residuals


Hi!

I am fitting a time-series model using proc arima in SAS.I fitted the
model with the AR,MA parameter and model is parsimonious with smalles
AIC,BIC and variance.In short looks like a best fit but unfortunately
I am not been able remove the correlation in the residuals I tried
with different parameters but correlation of residuals does not goes
away.
There is no exogenous variable available.
I tried series with over differencing and taking Log transform also
but the problem persists..
can anybody help me in how I can model this time series using proc
arima?
or any other suggestion is welcome.
Thanks in advance,
Sharad
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================



.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to