Mattias Villani: Spectral Subsampling MCMC for Stationary Multivariate Time Series

Abstract

I will present how a multivariate generalisation of the Whittle likelihood can be used to accelerate MCMC/HMC sampling for stationary multivariate time series by subsampling matrix-valued periodogram observations in the frequency domain. The methodology is illustrated and evaluated on a new multivariate generalisation of the autoregressive tempered fractionally integrated moving average model (ARTFIMA).