A1259
Title: Higher-order integer autoregression for count time series
Authors: Robert Jung - University of Hohenheim (Germany) [presenting]
Andrew Tremayne - University of Liverpool (United Kingdom)
Abstract: First-order integer-valued autoregressive models are widely regarded as a theoretical workhorse for the analysis of count time series. While the theoretical literature on these models is extensive, relevant applications in empirical work remain relatively scarce. Moreover, little attention has been paid to the consequences of choosing higher-order specifications. The aim is to examine the role of integer-valued autoregressions beyond first order, with particular emphasis on how dynamic properties, both first- and second-order, affect inference and forecasting. Using a combination of real-world data and simulation experiments, the implications of alternative model specifications are assessed. The analysis is facilitated by the newly developed R package coconots. The results show that the choice of lag order has non-trivial consequences, and that straightforward extensions of the basic first-order framework are not always appropriate.
