Proc esm

[parmis]

Hello,

I'm trying to forecast a variable for the next 36 months. my actual data has  seasonality but it is constant at the same time. I was using the following method to forecast for 12 month and I was happy with the result, but when I expand my lead time to 36 the numbers increase for the second year. Is there a way to modify the code to make the forecast constant? (the variable amount is always in the range of 100000-110000)

proc esm data=have out=want lead=36;

id date interval=month;

forecast amount/ model=addwinters transform=log;

run;

Thank you!

[parmis]

Hello,

I'm trying to forecast a variable for the next 36 months. my actual data has  seasonality but it is constant at the same time. I was using the following method to forecast for 12 month and I was happy with the result, but when I expand my lead time to 36 the numbers increase for the second year. Is there a way to modify the code to make the forecast constant? (the variable amount is always in the range of 100000-110000)

proc esm data=have out=want lead=36;

id date interval=month;

forecast amount/ model=addwinters transform=log;

run;

Thank you!

[imvash]

Your question is vague. The statement "my actual data has seasonality but it is constant at the same time" does not make sense to me. Do you mean there's no obvious increasing/decreasing "trend" in your data? How many historical observations do you have? In my opinion, the best practice is to have a holdout region to diagnose the best model among ARIMA and ESM by using HPFDIAGNOSE procedure, instead of fitting only "addwinters" esm model.

[parmis]

Yes, that's what I meant. My data is always in the range of 30000 to 45000, but there is a repeated monthly pattern. ex:the numbers drop every February, April and June.