Dear Allstat,
I have fitted a model to monthly trade data. Ultimately I will use the
monthly predictions to estimate quarterly data.
However, for the moment I want to see how well the model predicts by
estimating known monthly values and examining the residuals.
For example, I have real data up to Dec 2001 and I will run the model up
to Nov 2001 and predict Dec 2001 .
Ditto for Nov 2001 ( using Oct 2001 )
Ditto for Oct 2001 ( using Sep 2001 ) and so on.
This is tedious in the sense that I have to run the model and capture the
estimate ( and residual ) each time and then run the model again etc .
If this is a recognised diagnostic approach then is there a routine in
proc arima to do it automatically or has anybody written a macro to do it.
If this is successful , is there any advantage in repeating the exercise
for a lag of 2 or 3 etc ? I suspect not for if a model cannot predict
ahead 1 period then it would be pure spurious chance that it did for 2 or 3
periods in the future.
(Obviously I would see some evidence of the seasonality factor at the
appropriate lag )
Any comments , advice or bad experiences much appreciated !
John Milne
Balance of Payments Section
Central Statistics Office of Ireland
Rathmines
Dublin 6.
Tel : + 353 - 1 - 4984205
The information in this email, and any attachments transmitted with it, are confidential
and are for the intended recipient only. If you receive this message in error, please
notify us via [log in to unmask]
To see the latest figures from the CSO go to http://www.cso.ie
|