Economic Time Series and Seasonal Adjustment Methods

该项目的主要目的是重新审视日本中央和地方政府公布的经济时间序列数据中常用的现有统计方法。我们特别研究了美国人口普查局最近开发的X-12-ARIMA方法和统计数学研究所的北川教授开发的DECOMP方法。首先，我们调查了X-12-ARIMA方法的主要改进，该方法是Census X-11方法的修订版。由于X-11方法在日本政府官员中普遍使用，因此改进的意义一直是我们研究的中心问题。我们发现使用X-12-ARIMA方法通常可以得到稳定的时间序列数据集，但也发现这实际上取决于程序中使用的季节ARIMA模型的选择。另一个问题是，我们是否应该使用交易日调整和闰年调整来制作更正式的时间序列。由于交易日效应和闰年效应背后的时间序列周期不是季节性的（即12个月的周期），如果我们在X-12-ARIMA计划中使用这些选项，仍然存在争议。我们还研究了X-12-ARIMA程序和DECOMP程序残差的光谱特性。我们发现X-12-ARIiMA残差估计的光谱通常是平滑的，而DECOMP残差估计的光谱在季节周期中有时会出现下降。我们试图研究这种现象是否是在MSE意义上的DECOMP程序的最优性质的结果。这一问题是Grether和Nerlove在季节调整方法的经典研究中指出的，我们做了一些模拟研究。然而，在这个问题上我们还不能得出一个确定的结论。根据我们的调查，我们有一个印象，我们需要从理论方面和日本政府的实践方面对这两个季节调整计划进行更多的研究。总之，我们已经完成了这个项目最重要的目标。参与该项目的三名成员撰写了一些论文，也激励了大量相关领域的研究人员和日本政府的一些统计人员。我们感谢文部科学省对研究项目的慷慨支持。少

The main purpose of this project was to re-examine the existing statistical methods often used in making the published economic time series data from the central and local governments in Japan. In particular we have investigated the X-12-ARIMA method recently developed by the U.S.Census office and the DECOMP method developed by Professor Kitagawa of the Institute of Statistical Mathematics.First we have inverstigated the major improvements in the X-12-ARIMA method, which is a revised version of the Census X-11 method. Since the X-11 method has been commonly used among Japanese governrment officials, the meaning of improvements have been the central issues in our study. We found that we can often get stable time series data sets by using the X-12-ARIMA methods, but also found that it really depends on the selection of the seasonal ARIMA models used in the program. Another issue has been whether we should use the trading day adjustments and the Leap year adjustments in order to make the … More official time series. Since the time seresi cycles behind the trading day effects and the Leap year effects are not seasonal (i.e. 12 months cycles), it has been still controversial if we use these options in the X-12-ARIMA program. We also have investigated the spectral properties of the residuals from the X-12-ARIMA program and the DECOMP program. We found that the estimated spectrum from the X-12-ARIiMA residuals often are smooth while the estimated spectrum from the DECOMP residuals have sometimes dips in the seasonal cycles. We have tried to investigated if this phenomenon is the result of the optimal properties of the DECOMP program in the sense of MSE.This problem was pointed out by the classical study on the seasonal adjustment methods by Grether and Nerlove and we have done some Simulation studies. However, we could not have reach a firm conclusion on this issue. Given our investigations, we have an impression that we need more study on these two seasonal adjustment programs from the theoretical side as well as the practical side in the Japanese governments.In conclusion, we have acomplished the most important objectives of this project. Three members participated in this project has written some papers and also stimulated a large number of researchers in the related fields and some statisticians in the Japanese governments We thank The Ministry of Education, Science and Culture for giving the generous support to research project. Less