Announce of Seminar

Prof. Carlo Gaetan

Dipartimento di Scienze Ambientali, Informatica e Statistica Ca' Foscari University

A MODEL FOR TEMPORAL EXTREMES BASED ON LATENT PROCESSES

December 14th, 2011 at 10.30am Aula Consiglio - Ala 2C, S. Giobbe, Venezia

Anyone interested is invited to participate

One approach of modelling extreme data is to consider the distribution of exceedances over a high threshold. Under suitable conditions, this distribution can be approximated by a generalized Pareto distribution (GPD). In recent research on extreme value statistics, there has been an extensive development of threshold methods for time series models. For instance extreme values of univariate time series are modelled by assuming that the time series is Markovian and using bivariate extreme value theory to suggest appropriate models for the transition distributions. Another possible approach for dealing with the dipendence is a hierarchical approach imposing a prior distribution on the parameters. A drawback of the Bayesian approach is that the resulting marginal distributions are not GPD. In this talk we present a new model for the extremal behaviour of a univariate time series based on a latent process, that overcomes this drawback. The extremal properties of the model are illustrated, showing that different choices of the underlying latent process can produce different degrees of asymptotic dependence, from independent extremes to clustering of exceedances. Two analysis of environmental and financial time series are also presented.

(a joint work with Paola Bortot, Universitą di Bologna)

The announcement is available here