| Abstract [eng] |
The integer-valued autoregressive process is commonly used to model time-series data. One of the challenges for such models is dealing with overdispersion. In this paper a new integer-valued autoregressive model, denoted as BT-INAR(1) is introduced, which can be used for data that exhibits overdispersion. The model consists of binomial thinning operator and Bell-Touchard innovations. A number of BT-INAR(1) model's properties are derived, such as mean, variance, covariance, autocorrelation, and joint probability function. The model's parameters are estimated using Yule-Walker and Conditional Maximum Likelihood methods. Parameter estimates are compared via Monte Carlo simulation. Finally, the model is applied to two real time-series count data sets, and the model's performance is evaluated. |
