Abstract:
|
Sometimes the last 8-9 years of a monthly series have a well fitting ARIMA model whose goodness of fit diagnostics deteriorate greatly with the full data span. To obtain a better SEATS adjustment than that of X-13ARIMA-SEATS applied to the full series, we investigate a moving data-window approach somewhat like that of BAYSEA (1980) but enhanced with the automatic outlier treatment of X-13ARIMA-SEATS, for 9 year long data windows that advance a year at a time. The ARIMA model's differencing, AR and MA orders are determined from the last 9 years of data. The model's ARIMA and regression parameters are re-estimated for each window along with outlier re-identification. The estimated model for the middle year of the window defines the published adjustment when there are at least four preceding and four subsequent years in the full data span. The first and last windows define the adjustments for the first and last four years of the series. We examine the consistency of outlier identification where windows overlap, and the evolution of the model's coefficients, model quality diagnostics, and adjustment quality diagnostics to identify series for which this approach can be recommended.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.