On the novel geometric and negative binomial INAR(1) processes

TL;DR

This paper explores new geometric and negative binomial INAR(1) models based on a novel approach that specifies marginals and innovations first, then identifies thinning operators, connecting and extending previous work in the field.

Contribution

It establishes the existence of these novel INAR(1) models, strengthens previous results for the geometric case, and extends findings to the negative binomial case.

Findings

Connected the new models to existing thinning operators

Strengthened results for geometric INAR(1) models

Extended results to negative binomial INAR(1) models

Abstract

Guerrero et al. \cite{GBSO} propose a novel approach to building first-order integer-valued autoregressive (\inar1) models based on the concept of thinning. The standard approach requires that the thinning operator be defined first and \inar1 models with either a specified marginal (the forward approach) or a specified innovation (the backward approach) are developed. In contrast, the approach in \cite{GBSO} is to start out by specifying both the marginal distribution of the process and that of its innovation sequence, and then proceed to identify the thinning operator by solving a functional equation. In this article we discuss the connection between the thinning operators the authors obtained for their novel geometric and negative binomial \inar1 models and the thinning operator introduced in \cite{AB1} and \cite{AB2}. More specifically, we show that the existence of the two models has been established in \cite{AB1} using the forward approach and a different parameterization. In the process, we strenghthen some of the authors' results obtained for the novel geometric \inar1 process and we extend their results to the novel negative binomial \inar1 process.