Title: Bayesian spectral analysis of high-dimensional time series
Authors: Robert Krafty - University of Pittsburgh (United States)
Zeda Li - Baruch College (United States) [presenting]
Abstract: A frequency-domain factor model is proposed which allows for complex-valued spectra which means that individual high-dimensional time series can propagate in a lagged fashion. Our model allows for different dynamics across the variates of the time series. The spectrum of the factors is assumed smooth as a function of frequency. The real and imaginary parts of the loading matrix are modeled by tensor products. Inference is performed by MCMC methods, and the method is illustrated with biomedical data.