06-12-2016 05:45 AM - edited 06-12-2016 05:46 AM
I have used SAS forecasting studio for producing forecasts of a months sales. There I can set hold out period for identifying the best model based on desired measure such as MAPE, RMSE etc. Proc HPFdiagnose runs in the back end. Recently I am trying learn how to use proc arima /esm/ucm for forecasting. How do I set hold out period and best model selection criteria? Do I need to keep hold out period data aside separetely and forecast using insample data for those hold out period and manually measure the accuracyy of forecast?
06-15-2016 02:14 PM
SAS Forecast Server is designed to tackle large scale forecasting problems - as such you will find functionality in this product which is specifically designed for these kinds of challenges, such as picking a champion model for each time series at hand from a wide range of statistical forecasting models automatically.
The procedures you are referring to, such as ARIMA, ESM, and UCM, were implemented with a different purpose in mind, which is allowing a modeller to focus on a couple of time series only, and try to come up with the very best model, by fine tuning options manually.
Of course there is overlap between the 2 domains, and things are not as black and white as my earlier statement may suggest.
If you like to mimic SAS Forecast Server like behaviour with these procedures, you probably will have to write a lot of SAS code yourself.