A1225
Title: Bayesian Markov switching tensor regression for time-varying networks
Authors: Roberto Casarin - University Ca' Foscari of Venice (Italy) [presenting]
Monica Billio - University of Venice (Italy)
Matteo Iacopini - Vrije Universiteit Amsterdam (Netherlands)
Abstract: A Bayesian Markov Switching regression is proposed for modelling dynamic multilayer networks. We apply a suitable low-rank decomposition of the tensor of coefficients for parsimony and avoiding the over-fitting. The time-varying parameters are driven by a hidden Markov chain whose states are identified by means of constraints imposed on the regime specific parameters which determine the number of observed edges. In addition, we jointly model a vector of observables. We exploit the Polya-Gamma data augmentation scheme for logit models in order to provide an efficient Gibbs sampler for posterior inference. We show the effectiveness of the sampler on simulated and real datasets.
