Aggregation and Model‐based Methods in Seasonal Adjustment of Labor Force Series

Stuart Scott and Peter A. Zadrozny


One of the most closely-watched numbers produced by the U.S. Bureau of Labor Statistics (BLS) is the seasonally adjusted civilian unemployment rate. It is computed from eight employment and four unemployment series. A motivation for examining this issue of aggregation is that four of twelve components of the unemployment rate are for teenagers and four are agricultural employment, series with substantial sampling error. This study revisits work of Estela Dagum (1978).International interest in model-based seasonal adjustment has increased with availability of the TRAMO/SEATS software of Gomez and Maravall (1994). The methodology, which builds on the work of Burman (1980), carries out signal extraction on seasonal ARIMA models. Comparisons will be made between TRAMO/SEATS and X-11 methodology using Census's X-12-ARIMA (Findley et al, 1998).