Estimation and discrimination in locally stationary time series models

Wolfgang Polonik
University of California, Davis

(October 25, 2005 4:30 PM - 5:30 PM)

Estimation and discrimination in locally stationary time series models

Abstract

Locally stationary time series have formally been introduced by Dahlhaus (1997). A simple example is provided by an autoregressive process with time varying parameters. We show how such models can successfully be applied to the problem of discriminating seismographic readings of earthquakes and explosions. Discrimination is here based on functionals of estimated time varying variance functions. Our method has the advantage that no alignment of the underlying time series is required. Estimating the variance functions is accomplished via a minimum distance approach. We utilize prior knowledge about our target problem by introducing shape constraints into the estimation process. Some justification for our method in form of large sample results will be presented, and the method is illustrated using simulations and a real data application.

This presentation is based on joint work with G. Chandler and R. Dahlhaus.