The objective of the research was, based on the recent progress of computer capabilities, to systemize the environment for advanced time series analysis with various models, algorithms, computational methods and softwares. For this purpose, we performed the followings :
(1)Development of Generic Time Series Model and Related Computational Methods.
We developed a time series analysis method based on general state space model which can be applied to very wide class of nonstationary nonlinear or non-Gaussian time series models. We published many papers related to this subjects.
(2)Research on New Information Criteria
We investigated new information criteria EIC and GIC.EIC is based on bootstrap bias correction for the log-likelihood, On the other hand, GIG evaluates the bias for any estimators defined by statistical functionals, By these two information criteria, it becomes possible to evaluate and compare models whose parameter are estiamted by various methods. Further, we refined the bias correction by GIC and developed a improved version of the generalized information criterion.
(3)Development of Interfaces for Organization of Softwares
Recent development of computer networks such as internet makes it easy to distribute or access to newly developed time series analysis softwares, In this research, Sato developed a unified method based on the Web. In this method, since all of the computations are performed within the server computer, user can always use these sofiwared only if they have access to internet and brouwser, In this research, he forcus on the seasonal adjustment program DECOMP, On the other hand, Ishiguro developed a software for the analysis of multivariate system based on multivariate AR model.