Article ID: | iaor20053377 |
Country: | Netherlands |
Volume: | 20 |
Issue: | 4 |
Start Page Number: | 551 |
End Page Number: | 556 |
Publication Date: | Oct 2004 |
Journal: | International Journal of Forecasting |
Authors: | Findley David F., Wills Kellie C., Monsell Brian C. |
Keywords: | ARIMA processes, seasonality |
We examined the shrinkage methods of Miller and William from the perspective of seasonal adjustment rather than forecasting, restricting attention to their performance on the approximately 500 of the 1428 M3 series that are seasonal and have multiplicative seasonality. Local shrinkage improved the quality of the seasonal adjustment of enough of these series that almost 50% have acceptable automatic X-12-ARIMA adjustments, instead of 40% with no shrinkage. For a few series, global shrinkage produced demonstrably incorrect results, and for some of these series and also others improved by local shrinkage, the SEATS seasonal adjustment provided by an experimental version of X-12-ARIMA offered still greater improvements. No benefits were observed from global shrinkage.