Vladas Pipiras a Professor in the Department of Statistics and Operations Research at the University of North Carolina, Chapel Hill. His recent research has focused on modeling high-dimensional time series, extremes and uncertainty quantification in physical systems, and sampling and streaming algorithms in connection to “big data”.
Talk: Multivariate count time series modeling through latent Gaussian processes
Abstract: The focus of this talk will be on multivariate count (ordinal) time series modeling through deterministic functions of latent Gaussian series that bin continuous to discreet values. The construction is such that the deterministic functions ensure desired marginal distributions for counts in each dimension, depending on unknown parameters, while the vector Gaussian series provides flexibility to model temporal and cross-sectional dependencies, usually through some parametric model like vector autoregression (VAR). Several estimation methods for parameters of interest will be discussed, including the method based on relating the second-order moments of the count and Gaussian series, possibly with penalization in the high-dimensional regime. With the latter method and in the high-dimensional regime, sparse VAR transition matrices are proved to be estimated consistently. Some data applications will also be mentioned. The talk will cover parts of several recent papers of the speaker and co-authors.