Bayesian Spectral Analysis of Replicated Multiple Time Series with Multiple Covariates

• Activity Number: 421
• Type: Contributed
• Date/Time: Tuesday, August 11, 2015 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Ori Rosen* and Robert Krafty and Sally Wood
• Companies: The University of Texas at El Paso and Temple University and The University of Sydney Business School
• Keywords: MCMC ; Multivariate Time Series ; Smoothing Splines ; Spectral Analysis ; Whittle Likelihood

Abstract

We propose a frequency-domain model for fitting multiple time series collected on subjects along with covariates. The underlying spectral matrix is modeled as a function of both frequency and covariates using a mixture formulation. The multivariate Whittle approximation is used for approximating the likelihood, where the spectral matrix is expressed as a modified Cholesky decomposition. Smoothing splines are used for estimating the elements of the spectral matrix as a function of frequency. The proposed method is applied to sleep research data.