The aim of this work is to assess the modeling performance of two bivariate models for time series of counts, within the context of a forest fires analysis in two counties of Portugal. The first model is a periodic bivariate integer- valued autoregressive (PBINAR), easily interpreted due to the PINAR description of each component. The alternative model is a bivariate dynamic factor (BDF) that has a flexible structure, with the dynamics described through the mean value of each component that is a function of latent factors. The results reveal that BDF model exhibits a better ability to capture the dependence structure.

CEMAT - Center for Computational and Stochastic Mathematics