Title: On modelling multivariate time series of counts
Authors: Dimitris Karlis - RC Athens University of Economics and Business (Greece) [presenting]
Abstract: Multivariate count models appear in many disciplines, like marketing, epidemiology, seismology just to name few. In many cases the data are time series, i.e. multiple counts observed in a sequel of time points. Hence, for correct modelling we need to account for both serial and cross correlations. While the literature on continuous time series abandons, discrete valued time series have been given less interest. There is an increasing literature and number of models in recent years. For multivariate count time series thing are even more sparse. Observation driven models for the univariate case are now quite popular. We have previously introduced and extended such models in the multivariate case offering different approaches for model building and estimation. The aim is to introduce such models and provide some new results related to this family of models, including model selection and computational approaches to improve the performance of multivariate autoregressive models. Real data applications will be also described from different disciplines.