Research on the stability and robustness of seasonal adjustment procedure

一般状态空间模型应用于季节调节中的各种类型的问题。特别是蒙特卡罗滤波、平滑和自组织状态空间模型对于统计季节调整的稳定和鲁棒处理是有用的。准确地说，提出了乘法型非线性季节调整方法、小计数数据的季节调整方法和季节调整中异常值自动检测方法。此外，还探讨了模式平均型季节调整。换句话说，我们不局限于一个特定的季节模型，而是考虑和监测所有可能的季节模型，并通过加权这些模型来实现预测。另一方面，作为时间序列问题的推广，通过条件强度方法的点过程建模，考虑了去除高频金融数据的日内周期性。结果表明，常用的样条平滑法估计日时函数并不能完全消除这种日内周期性。在这项研究资助的最后一年，1月31日和2月1日在东京举行了一次关于统计季节调整的国际研讨会，题为“季节性和周期性建模”，并取得了巨大成功。新开发的软件E-Decomp以免费光碟形式发放给与会者。

General state-space models are applied to various types of problems in seasonal adj ustment. Especially, Monte Carlo filter, smoothing and self-organizing state-space model are found to be useful for stable and robust treatment of statistical seasonal adjustment. To put it precisely, new methods are developed for multiplicative type non-linear seasonal adjustment, for seasonal adjustment in small count data, and for automatic outlier detection in seasonal adjustment. Furthermore, model averaging type seasonal adjustment has been explored. In other words, we do not confine the seasonal model to a specific one but consider and monitor all the possible seasonal models, and realize prediction by weighting these models. On the other hand, as a generalization of time serics problem, removing intraday periodicity in high frequent financial data is considered via point process modeling by conditional intensity approach. It is shown that commonly employed method that use spline smoothing to estimate time-of-day function does not completely remove such intraday periodicity. In the final year of this research grant, an intemational symposium on statistical seasonal adjustment was held in Tokyo under the title 'Modeling Seasonality and Periodicity' on January 3 1 and February I, which ended in a great success. Newly developed software E-Decomp was released and distributed to the conference participants on free CD-ROM.