Reduced-Rank Methods in Multivariate Time Series Analysis

• Activity Number: 274
• Type: Invited
• Date/Time: Tuesday, August 5, 2003 : 2:00 PM to 3:50 PM
• Sponsor: Business & Economics Statistics Section
• Author(s): Gregory C. Reinsel*+
• Companies: University of Wisconsin
• Keywords: autoregressive models ; cointegration ; common features ; reduced rank

Abstract

Interest in multivariate time series modeling has been growing in part due to development and use of reduced-rank and other structured parameterization methods that help to alleviate the difficulties associated with the large number of parameters in multivariate models. The formulation of reduced-rank structure for multivariate time series models will be reviewed, with particular emphasis on nested reduced-rank vector autoregressive (VAR) models. Methods and properties of specification and parameter estimation of such models will be highlighted. The use of reduced-rank methods for the long run matrix in VAR models, to investigate issues of cointegration and common trends in unit-root nonstationary vector time series, will also be discussed. Properties of likelihood ratio tests for cointegration rank and of reduced-rank estimators in cointegration models will be surveyed. The use of reduced-rank methods in VAR models to identify and represent general cofeatures, such as codependence and common cycles in economic time series, will also be considered. Application of reduced-rank methods to a multivariate time series of economic variables will be presented for illustration.