Modeling \(\mathbb{Z}\)-valued time series based on new versions of the Skellam INGARCH model (Q2233662)

Point processes are used in spatial analysis, e.g., to describe the evolution of high frequency financial prices or seismic activity at a high level of quantitative detail, just to give a few examples. The observed data is integer-valued on \(\mathbb{N}\) or \(\mathbb{Z}\), e.g., financial prices measured in cents or the number of earthquakes. This paper deals with integer-valued GARCH (INGARCH) models with Skellam (Poisson difference) distributed innovations. Whereas classic GARCH models with innovations following the normal distribution cannot explain the stylized facts of high-frequency financial data arising from the discrete nature of the price grid, Skellam INGARCH models close this gap. The first contribution of this paper is to generalize the symmetric Skellam INGARCH model already proposed in the literature by allowing for non-zero mean. The second contribution is to introduce a modified Skellam INGARCH specification which performs better in the case of over- or underrepresentation of some specific integers ($-1$, $0$ and $1$) in the data. The paper provides simulation results on parametric estimation and inference performance and an empirical application with high-frequency financial data.