Seasonal Adjustment of Weekly Time Series with Application to Unemployment Insurance Claims and Steel Production
William P. Cleveland, Stuart Scott
Seasonal adjustment of weekly data poses special problems because the data are not exactly periodic. The workhorse programs X-12 ARIMA, TRAMO/SEATS, and STAMP, are not suitable. Harvey, Koopman, and Riani (1997) introduced a structural model in which the seasonal component is modeled nonparametrically via periodic splines. Pierce, Grupe, and Cleveland (1984) captured a deterministic seasonal component using regression on trigonometric series within an ARIMA framework. The method advanced here uses the same trigonometric components, but adopts a locally weighted regression to capture changing seasonality. The method is illustrated with unemployment insurance claims data published by the U.S. Bureau of Labor Statistics and steel production data. It is being used successfully for these series and for weekly money supply series at the Federal Reserve.
Unobserved components, weekly data