ABSTRACT
I. INTRODUCTION General state space modeling, initiated in Kitagawa [1], is a Bayesian modeling of not-necessarily linear not-necessarily Gaussian time series. It is an extension of our earlier work on “smoothness priors” , a Bayesian linear Gaussian modeling of time series. That work in smoothness priors (reviewed in [2]) was motivated by Akaike [3], a penalized likelihood constrained least squares computational approach. Kalman filter-type state space approaches to that modeling were introduced in [4] and [5]. The term “smoothness priors” , adopted from Shiller [6], was used in [7-9]. Subsequently we attempted to redo conventional time series analysis from the linear Gaussian smoothness priors approach as well as extend that approach to the analysis of previously unaddressed time series modeling problems including the modeling of multivariate nonstationary covariance time series.
