Modeling and analysis of count data time series: recent research activities
 
 
Description:  After a brief introduction to basic concepts and examples of count data time series, we shall focus on two particular models and their statistical analysis: the (compound) Poisson INAR(1) model and the (self-exciting threshold) binomial AR(1) model.
The Poisson INAR(1) model is the most simple approach to model time series of (equidispersed) counts with an infinite range. We consider the compound Poisson INAR(1) model as a very flexible extension, which allows the counts to show overdispersion, and which also includes some other known instances of INAR(1) models as a special case. We discuss properties of this model (like stationarity, mixing) and ways of numerically computing the marginal distribution. Then we look at certain diagnostic test statistics, which are derived from dispersion, skewness or time-reversibility properties of INAR(1) processes. We derive closed-form expressions for their asymptotic distribution and present some results concerning the performance of the resulting tests.
If the range of the time-dependent counts is finite with a fixed upper limit, then the binomial AR(1) model is a popular choice for describing such data. After a review of some recent results with regard to this model, we propose the self-exciting threshold binomial AR(1) model as an extension allowing for piecewise-type patterns. Properties of this model are studied, and approaches for estimation and forecasting are considered.
Date:  2015-02-03
Start Time:   14:30
Speaker:  Christian H. Weiß (Helmut Schmidt Univ. Hamburg, Germany)
Institution:  Helmut Schmidt Univ. Hamburg, Germany
Place:  Room 5.5
Research Groups: -Probability and Statistics
See more:   <Main>  
 
© Centre for Mathematics, University of Coimbra, funded by
Science and Technology Foundation
Powered by: rdOnWeb v1.4 | technical support