Abstract
Time series count data such as daily cases of Covid-19 requires adequate modelling and
forecasting. Traditional time series models do not have limitations in modelling time series
count data, also known as unbounded N-valued data. This study involved in-depth analyses
of various models in fitting time unbounded N-valued data. Models such as the Zero-
Inflated Poisson, zero-inflated Binomial, and ARIMA popularly used to fit time series
count were compared with the integer-valued generalized autoregressive conditional
heteroscedasticity (INGARCH) models. The investigation involved two critical aspects:
simulation and real-life data analysis. First, we simulated the time series count data,
modelled and compared the performance of the competing models. The simulation
outcomes consistently favoured the Negative Binomial INGARCH models highlighting
their suitability for count data modelling. Subsequently, we examined life data on Covid-
19 data in Nigeria. The life data also yielded strong support for the NB INGARCH model.
This study recommends further exploration of the NB INGARCH model, as it exhibits
substantial promise in effectively modelling over-dispersed zero-inflated data. The current
study contributes valuable insights into selecting appropriate models for time series count
data, addressing the intricate challenges posed by this specialized data type. Also, the
overall outcome of the study helps in national planning, and resource allocation for the
people needing health intervention.