In SAS High-Performance Forecasting^® 12.3, changes were made to the seasonality testing routine used by the HPFDIAGNOSE and HPFENGINE procedures. These changes were intended to make it more likely that a seasonal model would be selected for a series, as compared with previous releases of the software--particularly for weekly series, where the seasonal index is large (that is, 52).

After the implementation of the new algorithm, a problem was found in the AIC calculations used to determine whether a series is deemed to be seasonal. This problem in the AIC calculation might prevent seasonal models from being considered, even if a series exhibits seasonal variation. In some cases, series that had been deemed to be seasonal in earlier releases of the software are now being deemed nonseasonal by the new algorithm. This problem impacts not only the SAS High-Performance Forecasting procedures, but also SAS Forecast Server^® 12.3, which uses the underlying SAS High-Performance Forecasting procedures.

To circumvent the problem in either the SAS High-Performance Forecasting or SAS Forecast Server products, provide your own model repository that contains the full set of models that you want the software to consider, including any seasonal models.

Operating System and Release Information