topic Re: Proc ESM forecasting - Multi-step? in SAS Procedures

Proc ESM forecasting - Multi-step?

Sroy9 — Tue, 15 Mar 2011 04:27:43 GMT

Hi,

I'm a bit new to using SAS, but am currently attempting to use it to perform Winter's forecasts on a set of data. I found that in using the ESM procedure the parameters are optimized, which is great, but I am having some trouble specifying multi-step ahead forecasting. I know that setting lead will add additional forecasts to the end, but I want that process to occur throughout.

For instance, if I set a lead time of 2 and have 200 data points, the forecast for data point 150 should be set on 148, and remain the same from then on. In this way I have a list of forecasts that are all formed 2-periods out from the original data set.

If anyone could point me in the right direction I would greatly appreciate it. Message was edited by: Sroy9

Re: Proc ESM forecasting - Multi-step?

udo_sas — Tue, 26 Jul 2011 15:46:01 GMT

Hello -

Provided I understand correctly what you would like to archive, I think the answer is to run ESM several times using the BACK= option and to merge the results of each run.

Here is an illustration of what I'm referring to: I running ESM 12 times with a lead time of one while decreasing the out-of-sample size for each run and keeping the prediction of each run.

Hope that helps.

Thanks,

Udo

%macro esm(value);
%do i=&value. %to 1 %by -1;
   proc esm data=sashelp.air out=_null_ outfor=work.outfor&i. lead=1 back=&i.;
     id date interval=month;
     forecast air / method=winters;
   run;
   data outfor&i.;
     set outfor&i.(keep=actual predict date) end=last;
     if last;
   run;
%end;
data result;
   set outfor1-outfor12;
   by date;
run;
%mend;

%esm(12)