topic Re: SES models in Proc ESM and Proc Forecast (using same level weight estimate) had different predic in SAS Forecasting and Econometrics

SES models in Proc ESM and Proc Forecast (using same level weight estimate) had different prediction

shenghua_ni — Fri, 05 Nov 2021 15:25:14 GMT

proc esm data=forecast  outfor=output1  print=all;
id sas_yyyymm interval=month; 
forecast TotalSpending;
run;

proc forecast data=forecast  interval=month 
method=expo weight=0.001 trend=1 
out=out outfull outresid outest=est;
id sas_yyyymm;
var TotalSpending;
run;

Use the same data set.
Use Simple Exponential Smoothing model in ESM and FORECAST.
The ESM gave the level weight esitmate=0.001.
I plug this as weight in FORECAST.
But the predict results from those two procedures are different.

Why the same SES with same weight got the different prediction?

Re: SES models in Proc ESM and Proc Forecast (using same level weight estimate) had different predic

sbxkoenk — Fri, 05 Nov 2021 16:19:00 GMT

Hello,

Which version of SAS are you in?

Submit:

%PUT &=sysvlong4;

to find out.

In my SAS 9.4 M7 (9.4 Maintenance Level 7) PROC FORECAST is still there in the software, but it is no longer in the accompanying documentation.

I know the doc was saying since a long time that the FORECAST procedure was obsolete.
Now I cannot even locate the doc for FORECAST procedure anymore (supporting the hypothesis that it is really obsolete!!).

Thanks,

Koen

Re: SES models in Proc ESM and Proc Forecast (using same level weight estimate) had different predic

shenghua_ni — Fri, 05 Nov 2021 18:44:48 GMT

SAS Viya V.03.05.

Re: SES models in Proc ESM and Proc Forecast (using same level weight estimate) had different predic

sbxkoenk — Fri, 05 Nov 2021 19:40:21 GMT

Hello,

The last documentation for PROC FORECAST can be seen for :

SAS 9.4 / Viya 3.3.

As from

SAS 9.4 / Viya 3.4
, there's no more documentation for PROC FORECAST.

You are on VIYA 3.5.

Do you absolutely need PROC FORECAST?
It's very sub-optimal compared to ESM procedure (the latter is optimizing smoothing weights, the former is not).
Even with same smoothing parms, differences can be due to missing value handling, forecast initialization and so on ...

Thanks,

Koen

Re: SES models in Proc ESM and Proc Forecast (using same level weight estimate) had different predic

shenghua_ni — Sat, 06 Nov 2021 00:51:48 GMT

No missing value.

Re: SES models in Proc ESM and Proc Forecast (using same level weight estimate) had different predic

sbxkoenk — Sat, 20 Nov 2021 20:32:49 GMT

Hello,

Are you still struggling with this?

If there are no missing values I suppose it has to do with the backcast initialization

(i.e. the number of observations used to initialize the backcast states).

But again, consider PROC FORECAST as obsolete.

Koen