10-26-2015 11:46 AM
I'm trying to figure out the technical processes for events in forecast studio. I assumed that they would operate in the model like dummy variables, but I've done a test to confirm this and it seems it is not so. I'm attaching a picture where on the left I use a level shift event and on the right I use a dummy variable that 0 before the event and 1 after. Everything else is specified the same. The differences between these two models largely relate to the CI associated with the forecasts (although the forecast estimates differ as well). However, when I do the same exercise and difference the dependent variable (and also difference the even and adjust the dummy variable accordingly) the differences are much more drastic. Can anyone comment on this issue and/or provide technical documentation of how events are functioning in the model?
10-26-2015 03:46 PM
i realized that an intercept was included in one and not the other....
i would still be very interested to see technical documentation for events, especially pulse or temporary changes if anyone knows where i can find it!
10-29-2015 12:12 PM
SAS Forecast Studio takes advantage of a procedure called HPFEVENTS for event modeling.
You will find HPFEVENTS documentation in the SAS Forecast Server Procedures 14.1: User's Guide (Chapter 9).