Spring 2001 Seminar Series - February 9, 2001
University of Minnesota
School of Statistics
College of Liberal Arts

Bayesian time-varying autoregressions: Models and Applications

Raquel Prado
Universidad Simon Bolivar
(Statistics Search Candidate)

February 9, 2001
4:00 PM, 110 Ford Hall
Minneapolis, East Bank Campus
Social at 3:30 PM, 300 Ford Hall

Abstract

Time-varying autoregressive (TVAR) models have proven useful in describing the behavior of long time series that experience changes in quasi-periodic content over time. Beginning with the class of TVAR models in Bayesian dynamic linear modeling framework, we review methodological aspects of time-domain decompositions that provide inferences on the structure underlying non-stationary time series. Recent model extensions that deal with model order uncertainty, and are enabled using Markov Chain Monte Carlo simulation methods, are discussed. We emphasize the relevance of TVAR modeling in the context of biomedical signal processing, in particular, we consider analyses of multiple electroencephalographic (EEG) traces that arise in various neurophysiological settings. We conclude with comments about current research on multivariate autoregressive models.