Exponential-family INGARCH models

Abstract

A myriad of integer-valued generalised autoregressive conditional heteroscedastic (INGARCH) models have been proposed in the literature, including those based on conditional Poisson, negative binomial, and Conway-Maxwell-Poisson distributions. This talk looks at a larger class of exponential-family INGARCH models, showing that maximum empirical likelihood estimation over this infinite-dimensional class can lead to consistent estimates and unbiased inferences on model parameters. The proposed framework is demonstrated on two datasets.