Research on Seasonal Adjustment of Economic Time Series

点击翻译按钮获取中文摘要

The objective of this research was to render the theoretical considerations to the various problems in statistical seasonal adjustment, and to develop new methods based on the state-of-art techniques in modern time series analysis. The summary of our research report follows.(1)Development of new proceduresSeasonal adjustment methods based on Monte Carlo filter and smoother, and on dynamical system approach were proposed. Dynamic X-11 model successfully gives explicit model form to X-11, which is defined in an essentially non-parametric way. Monte Carlo filter/smoother deals with higher-dimensional seasonal adjustment problem allowing non-linearity and non-Gaussianity, which was next to impossible by the existing statistical methods. Thsi method was applied to the seasonal adjustment of small count data which seemingly contains quasi-periodic components.(2)Enhancement and characterization of seasonal modelsMacroeconomic system was estimated based on the multivariate seasonal adjustment. These kind of composite seasonal adjustment could be more important in the near future. Extension of seasonal models to time-space data was also considered. Finally, the research on the characterization of linear Gaussian model based seasonal adjustment was done in terms of the optimality of seasonal adjustment. Comparative study between DECOMP and X-12 ARTMA was also performed through simulations.(3)Software developmentThe most successfull software product is Web-DECOMP.Users are free from instaling softwares on local disks or free from recompilation of software codes, but they simply can execute seasonal adjustment program DECOMP through internet browsers. On the other hands, U.S.Bureau of the Census almost completed their new version of software, X-12-ARIMA.The most updated version and its documentation (but not the official release yet at present) can be downloaded via their ftp site (ftp.census.gov).