16:00, January 24, Kennedy Seminar Room 2
Abstract: Instead of modelling Markov chains with discrete counts in time-series generalised linear model, we consider the application of tensor factorisation on the chain where its true order and internal serial dependence are unknown. The methodology is founded on Bayesian nonparametrics and based on conditional tensor factorisations, which helps to capture higher order interactions among the lags and then, maximal orders. Markov chain Monte Carlo is used to sample from the posterior distributions. We utilise some simulation experiments and examples to illustrate that our model works.