Title: Semiparametric observation-driven models for time-series of count
Authors: Thomas Fung - Macquarie University (Australia) [presenting]
Abstract: The aim is to show that observation-driven time-series models, such as the popular integer-valued generalized autoregressive conditional heteroskedastic (ingarch) and generalized linear autoregressive and moving average (glarma) frameworks, can be fit without correctly specifying the family of conditional distributions for the responses. Instead, the underlying family of conditional response distributions is treated as an infinite-dimensional parameter to be estimated from the data simultaneously with the usual finite-dimensional parameters. A key feature of this semiparametric approach is that it can automatically handle over and underdispersion, as well as other deviations from parametric models. Numerical studies suggest that both estimation and inferences using the approach can perform as well as correctly-specified parametric models, but can be more robust under model misspecification. Examples are given to illustrate the practical use of the methods.