ABSTRACT
Prior to the mid-1980s, a number of methods were developed to approximate the filtering/smoothing distribution for non-normal or nonlinear state-space models in an attempt to circumvent the computational complexity of inference for such models. With the advent of cheap and fast computing, a number of authors developed computer-intensive methods based on numerical integration. For example, Kitagawa (1987) proposed a numerical method based on piecewise linear approximations to the density functions for prediction, filtering, and smoothing for non-Gaussian and nonstationary state-space models. Pole and West (1989) used Gaussian quadrature techniques; see West and Harrison (1997, Chapter 13) and the references therein.
