A0757
Title: An observation-driven generalized Poisson model
Authors: Roberto Casarin - University Ca' Foscari of Venice (Italy)
Giulia Carallo - Ca' Foscari University of Venice (Italy)
Dario Palumbo - University Ca Foscari of Venice (Italy) [presenting]
Abstract: Based on the generalized Poisson (GP) conditional distribution, a new general class of observation-driven models for count data is presented, and their theoretical properties are derived. The GP is a flexible distribution that allows for both under- and over-dispersion. As a special member of this class, a score-driven model version is introduced. It is shown that this specification is robust to the presence of outliers and can be extended to allow for time-varying over-dispersion. For the estimation of the model, a Bayesian inference framework and an efficient posterior approximation procedure based on Markov Chain Monte Carlo are provided. Posterior contraction rates are also established with increasing sample size in terms of the average Hellinger metric. The applications on environmental variables show that the proposed model is well-suited for capturing the over-dispersion feature of the data.
