A0691
Title: Bayesian modeling of time series of counts under censoring
Authors: Isabel Pereira - University of Aveiro (Portugal) [presenting]
Abstract: Censored time series arise when explicit limits are placed on the observed data and occur in several fields including environmental monitoring, economics, medical and social sciences. The censoring may be due to measuring device limitations, such as detection limits in air pollution or mineral concentration in water. Censoring may also occur when constraints or regulations are imposed, such as in international trade studies where exports and imports are subject to trade barriers or hours worked, often treated as censored variables. The time series of counts under censoring are considered, focusing on the Poisson first-order integer-valued autoregressive(PoINAR) models. This class, while being simple and flexible, is useful for modelling positive-valued and integer-valued time series possessing an autoregressive structure with a non-negative serial correlation. Two natural approaches are investigated to analyze censored PoINAR(1) time series under the Bayesian framework: the Approximate Bayesian Computation (ABC) methodology and the Gibbs sampler with Data Augmentation (GDA) Approach. The parameter estimation performance of both approaches is made through a simulation study.
