Integer-valued autoregressive models with survival probability driven by a stochastic recurrence equation

TL;DR

This paper introduces a novel class of integer-valued autoregressive models where the survival probability is governed by a stochastic recurrence equation, enhancing model flexibility and performance in time series analysis.

Contribution

It proposes a new model framework with a stochastic recurrence-based survival probability and proves estimator consistency under misspecification.

Findings

Model improves in-sample fit for crime report data

Model enhances out-of-sample predictive accuracy

Simulation confirms flexibility of the proposed specification

Abstract

A new class of integer-valued autoregressive models with dynamic survival probability is proposed. The peculiarity of this class of models lies on the specification of the survival probability through a stochastic recurrence equation. The estimation of the model can be performed by maximum likelihood and the consistency of the estimator is proved in a misspecified model setting. The flexibility of the proposed specification is illustrated in a simulation study. An application to a time series of crime reports is presented. The results show how the dynamic survival probability can enhance both in-sample and out-of-sample performances of integer-valued autoregressive models.