forecasting Procedure for multiple seasonality

[Forecaster]

I have a forecasting task at hand that has multiple seasonality. As an example, hour of the day, day of the week, week of the month, and month of the year. Is there a SAS forecast procedure that can handle multiple seasonality, I have access to SAS forecast server, SAS /ETS. I tries using Proc UCM but it is very slow.

Thanks for your help, much appreciated

[stat_sas]

What is the time interval for your data like daily,monthly, yearly?

[Forecaster]

hourly.

[stat_sas]

After plotting your hourly data, how do you determine daily, monthly, yearly seasonality?

[Forecaster]

Visual inspection and business domain knowledge.

This is common in Call center activity data and electricity demand data. @stat@sas do you know of a procedure in SAS that would address multiple seasonality.

[stat_sas]

First I am trying to understand how hourly plotted data on the graph can have multiple seasonality factors. If there is some seasonality in the data it should be hourly not daily, montly or daily unless there is some method which can sum up 24 hours to one day, 7 days to one week etc. on the graph which does not seem to be a possibility. May be other experts can suggest some solution for this sort of problem.

[udo_sas · Accepted solution]

Hello -

UCM can be slow with hourly data due to the size of the state vector.

Alternatively you may want to break up the problem into several modeling runs - one for each seasonality pattern which you have in mind - and then reconcile these different forecasts of different frequencies using the HPFRECONCILE procedure (assuming you have access to SAS Forecast Server).

Thanks,

Udo