Abstract: Panel vector auto-regressive (VAR) models are effective in modeling the evolution of multivariate time series (with an identical set of variables) across different sub-populations. In this talk, I will introduce a novel panel VAR model with a shared low-rank structure modulated by sub-population-specific weights, and enhanced with idiosyncratic sparse components. Parameter identifiability issues are addressed through constraints that lead to a nonsmooth, nonconvex optimization problem. For parameter estimation, a multi-block ADMM algorithm will be presented along with its convergence properties under mild conditions. I will also touch on the consistency of the parameter estimates under high-dimensional scaling. The talk will conclude with a performance summary of the posited model, evaluated through simulations on synthetic data and an application to a neuroscience data set.

Presenter: Yuchen Xu
Affiliation: UCLA

Host department: Statistics

Location: E2-194