B1865
Title: Generalised dynamic factor models for high dimensional spatio-temporal random fields on a network
Authors: Chao Zheng - University of Southampton (United Kingdom) [presenting]
Abstract: High dimensional datasets containing records of spatio-temporal structures are of interest in many applications e.g., brain imaging, meteorology, and marketing research. We consider a dimensionality reduction technique using the generalized dynamic factors models. Different from the conventional approaches in time series where representation theory by using Wold's theorem and the concept of "innovations" is very natural, we have to take into account the spatial dependence where there is no unique definition of the concept of ``innovation''. To this end, we derive our generalized factor model by working on the spectral theory of random fields on 3D-grids. We established the rigorous definitions of multivariate space-time processes, its spectral representation theory, and a consistent spectral density estimator. Numerical and real-data examples are also provided.
