Development of Time Series Analysis Software Based on State-Space Modeling

点击翻译按钮获取中文摘要

By using the state-space model, most of the familiar time series models can be treated. It also provides efficient recursive filtering algorithm for state estimation, prediction, smoothing and parameter estimation based on the likelihood. In this project research, we developed a time series analysis software based on this state-space approach. Further, we developed a Web-version of the softwares and applied the method to various real-world problems. The major outcomes are as follows:1. Development of filtering and parameter estimation method.We refined the algorithms of the Monte Carlo filter and the Gaussian-sum filter for efficient computation of conditional state distributions. We also developed parallel algorithm for computing estimation of the state and some algorithms for simultaneous estimation of the state and the unknown parameters.2. We developed generic softwares for Monte Carlo filtering and self-organizing filter. Specific application softwares were also developed for trend estimation, estimation of stochastic volatility, trend and volatility model and power contribution analysis. Web-software for the application software on user's Web site without installing these softwares.3. The developed methods and softwares were applied to various real-world problems such as the seismology, earth-magnetography, finance, macro-economics, marketing, human behavior, ship engineering, and various other complex systems and obtained useful results in prediction, control and knowledge discovery.