ABSTRACT
Statistical agencies routinely seasonally adjust large numbers of time series generated from periodic labor force surveys. While these surveys produce highly reliable estimates for national aggregates, the national samples are spread too thin geographically to produce reliable subnational estimates. Moreover, these surveys often use a rotating panel design in which a portion of the sample is retained each period. As a result, survey error (SE) with complex autocorrelation patterns can be a major source of variation in the observed series. The purpose of this paper is to examine the effects of seasonally adjusting a labor force survey series with two widely used methods that ignore SE, X-11/X-12-ARIMA (Dagum 1980; Findley et al. 1998) and SEATS (Go´mez and Maravall 1994), and to show how adjustments produced by these methods can be modified to account for SE. The two methods differ in how signal extraction filters are obtained. SEATS filters are more tailored to the specific properties of the series than the X-11 filters. A question to explore is whether this difference in filter selection makes SEATS more effective in handling the
K12089 Chapter: 6 page: 135 date: February 14, 2012
K12089 Chapter: 6 page: 136 date: February 14, 2012
Modeling and
effects of SE than X-11. Where SE is small, the seasonal adjustments ignoring SE differ little from those that account for it.
