topic Custom model fails in Forecast Studio in SAS Forecasting and Econometrics

Custom model fails in Forecast Studio

kdp — Thu, 18 Feb 2010 16:03:55 GMT

Hello,

I am using Forecast Studio 2.1.

I created several fully-fitted regression model specs in E-Guide and I am trying to use them alongside the ARIMA and ESM models in Forecast Studio.

When I run the project, the ARIMA and ESM models work fine. However, every one of my custom regression models fail. I have no idea why and I don't know where to look.

I have used the same dataset to run the Linear Regression wizard in E-Guide and it works fine, so I know the specs I created should work.

Could anybody provide ideas on why the custom model fails in Forecast Studio?

Thanks,
kdp

Re: Custom model fails in Forecast Studio

udo_sas — Fri, 19 Feb 2010 07:51:24 GMT

Hello -
To view the SAS log that SAS Forecast Studio creates when generating the forecast, select Tools -- Log. This might give you a first idea what is going on.

As an idea: if you are using EG to fit regressions models - you are probably running the REG procedure. Note that since this is not a time-domain based procedure it is not supported as a custom model in SAS Forecast Studio.

Having said this, there are ways to fit multiple regression models in SAS Forecast Studio. What it does, though, is running an ARIMA model - leaving out all "ARIMA specfic components" like p,d,q and P,D,Q.
For example:
PROC HPFARIMASPEC
/* Label: Y = CONST + price*/
MODELREPOSITORY = work.temp
SPECNAME=MULTIPLEREGRESSION1
;
FORECAST TRANSFORM = NONE ;
INPUT SYMBOL = price
TRANSFORM = NONE ;
ESTIMATE
METHOD=CLS
CONVERGE=0.0010
MAXITER=50
DELTA=0.0010
SINGULAR=1.0E-7 ;
run;

Hope that helps - you might consider contacting Technical Support of course.

Regards,
Udo

Re: Custom model fails in Forecast Studio

kdp — Mon, 22 Feb 2010 15:14:12 GMT

Thanks for your suggestions Udo!

I followed your example and created the following spec and even this one fails:
***************************************
Proc HPFARIMASPEC
MODELREPOSITORY = ldp.kdp_rep
SPECNAME=FS_BASEARIMA
SPECLABEL=""
SPECTYPE=ARIMASPEC
SPECSOURCE=FSUI
;

FORECAST SYMBOL = value TRANSFORM = NONE;

INPUT SYMBOL = FS_Arr TRANSFORM = NONE;
INPUT SYMBOL = WH_Arr TRANSFORM = NONE;
INPUT SYMBOL = FSWH_Arr TRANSFORM = NONE;

ESTIMATE
METHOD=CLS
CONVERGE=0.0010
MAXITER=50
DELTA=0.0010
SINGULAR=1.0E-7 ;
run;
***************************************
Note: My dependent variable is called "value" and "FS_Arr", "WH_Arr", and "FSWH_Arr" are three of my independent variables in the dataset.

It's hard to know why this model fails as well since there are no errors reported in the log.

Thanks,
kdp

Re: Custom model fails in Forecast Studio

kdp — Mon, 22 Feb 2010 16:37:33 GMT

Just figured out why it was failing.

The "FS_Arr" value had zeroes throughout...taking it out of the spec fixed the problem.

However, not sure why this is causing an issue? Shouldn't it just come back with a coefficient of 0?

Re: Custom model fails in Forecast Studio

udo_sas — Tue, 23 Feb 2010 09:25:38 GMT

Hello -
Frankly speaking I don't know what this is causing the problem. At this stage I would recommend considering to open a track with Technical Support.
Thanks,
Udo

Re: Custom model fails in Forecast Studio

kdp — Tue, 23 Feb 2010 20:39:52 GMT

Thanks Udo! Will look into it.
kdp