Abstract:
|
The U.S. Census Bureau estimates "value of construction put in place" (VIP) each month using a variety of sources, including sample surveys, administrative records, and trade association data. There are plans to expand the publication of the VIP to more levels of detail, but some of the resulting direct estimates will have large sampling errors. This paper examines model-based time series smoothing techniques to reduce sampling error for both unadjusted and seasonally-adjusted estimates. The work involves the following steps: 1.) estimation of sampling error autocovariances and autocorrelations of the direct estimates; 2.) use of these to develop sampling error models; 3.) fitting models to the time series of direct estimate that include both sampling error components and components representing the true VIP series; 4.) application of signal extraction results to borrow information over time to improve estimates of the underlying true VIP series; 5.) application of model-based seasonal decomposition (Tiao and Hillmer 1982, Burman 1980) to the model for the true series and use signal extraction to produce seasonal adjustments with reduced effects of sampling error.
|