Abstract
Statistical modelling of count time series is challenging. In this talk we propose a class of models for time series of count data that allow a flexible treatment of time dependence as well as non-stationarity. The models are derived by exploiting the equivalence between discrete infinitely divisible and compound Poisson distributions. A Poisson point process and a time-moving trawl set are used to introduce time dependence. In the stationary setting, the trawl set is defined by probability density functions that capture the underlying dependence structure in a computationally efficient manner. Furthermore, we develop regression models by allowing the geometry of the trawl set to depend on covariate information, giving rise to rich dynamics. Our particular specification strikes a balance between tractability and the ability to capture complex dependence structures. Statistical inference and prediction is based on the bivariate probability function. The practical relevance of our models is illustrated by two real data examples in the fields of crime analysis and epidemiology. This a joint work with Bonface Miya Malenje (JKUAT, Kenya) and Thomas Opitz (INRAE, France).
Organizzazione
Monica Chiogna