03-17-2014 10:49 AM
I had a few questions about Forecast Studio 12.1, and was wondering if anyone on here had the answers.
1. Is it possible to select a model based on multiple selection criteria rather than just one - either through code or point and click options?
2. How does SAS detect seasonality and trend? What is the statistical test being performed?
03-18-2014 02:42 PM
1. No, commonly you select the accuracy statistic you are interested in first and then pick the winning model. Forecast Studio allows you to see all accuracy statistics and all models, though. Go to "Modeling View" and select: "Compare Models" icon. You should see a chart like this (note "Statistics of Fit" tab):
2. Please check out the SAS Forecast Server Procedures 13.1: User's Guide - Chapter 19 "Forecasting Process Details".
03-20-2014 06:17 PM
Got it, thanks for the response Udo!
A quick follow up question - something else that came up. What are the relative errors in the statistic of fit? Is this relative to a random walk model? i.e. Mean relative Absolute Error.
03-21-2014 04:33 PM
Relative refers to the relative absolute error (RAE). You will find an definition of all accuracy statistics in the procedure user's guide - check out "Statistics of Fit" in the Forecasting Process Details chapter.