Computational Methods for Time Series Analysis
Genshiro Kitagawa, Tomoyuki Higuchi, Seisho Sato · COMPSTAT · 2002
By the progress of fast computing facilities and various computing technologies, it becomes realistic to apply computer intensive methods to statistical analysis. In time series analysis, sequential Monte Carlo methods was developed for general state space models which enables to consider very complex nonlinear non-Gaussian models. In this paper, we show algorithms, implementations and parameter estimation for Monte Carlo filter and smoother. Various ways of the use of parallel computer are also discussed. The usefulness of the general state space modeling is illustrated with several examples.