10-21-2015 01:41 AM
I am working on AR model where in i found an issue. The code is as follows
proc autoreg data=sas_dataset;
model y= dummy1 dummy2 dummy3 dummy4
/method=ml nlag=12 garch=(p=1,q=1) maxiter=500 dwprob;
output out=Forecast p=yhat;
if I run the code in SAS9.4 the coefficients are different from the coefficients obtained in SAS EG 5.1.
If I remove the maxiter option in the code and run in SAS EG 5.1 then the coefficients are matching to some extent.
Why this descrepancy if any body can help me it would be great.
10-21-2015 09:47 AM
I think this would be question for SAS tech support. For your question itself, EG vs Base is like Apple vs. Orange. EG is a GUI(more or less) to foundation SAS. The difference you saw lies in between possible SAS/ETS versions, and a possible result of different Hardware/software setup. Here are some information you want to have when you contact SAS tech support.
1. Run proc setinit on both platforms to figure out the versions of both SAS/ETS. Record their site numbers.
2. The location of your SAS running may not be local. You need to know the hardware (CPU, memory etc) and OS (Windows/linux version etc) for both platform. EG could be running on the remote server without your knowing it, Base SAS need to run some special code to use remote application server (via metadata server or direct SAS /connect), but it could be also prepared in the config file or autoexec file , so you may not realize its whereabout either.
10-24-2015 06:06 AM
I am sorry!
The answer is just useless. How can the estimate get changed from version to version. I had been facing this severe issue for long time. Yet no solution I am getting. If there is something to do with seed then it could have made sense. Even there also we need to know the way to figure out the common seed so that we can get the consistent estimates irrespective of SAS version.
Again let me make this point clear. Version cannot change the formula for estimates.
If 2+2 =4 then it will be 4 only not 5 or 3 in different versions of calculator. Kindly introspect and try to post the firm response.
10-24-2015 10:44 PM
Make sure you are comparing apples to apples.
Finally make sure you understand the behavior of the procedure. Is it a deterministic calculation or does the procedure something such random sampling or random starting values that would cause two rins on the same computer to be different.
If you can reproduce the different results then open a support ticket with SAS. They can tell you exactly what is happening.