[Forum SIS] Seminari di Statistica - Genova - 5 aprile

[Maria Piera Rogantin]

Ho il piacere di invitarvi ai seguenti due seminari che si terranno 
      Giovedi 5 aprile con inizio alle ore 14
nell'aula 705 del Dipartimento di Matematica dell'Universita` di Genova. 
________________________________________________________

Speaker: Beniamino Hadj-Amar, Department of Statistics, The University of Warwick (UK) 

Title:  Simultaneous Bayesian model search and change-points detection for non-stationary oscillatory processes

Abstract: The analysis of time series that show oscillatory and - more or less - periodic behaviour is a common task in many sciences. In particular, the non-stationary analysis of cyclical data that show regime shifts in periodicity, amplitude and phase is challenging as the timing and number of such changes is usually unknown. We propose a methodology for detecting such regime shifts using several Bayesian transdimensional Markov chain Monte Carlo algorithms which allow for model searches between parameters subspaces of different dimensionality. Our algorithm incorporates multiple model searches, namely the number of regimes and the harmonic models that capture the relevant frequencies along with their amplitudes and phases within each regime. We illustrate the use of the methodology for data from experiments on rodents to detect instances of sleep apnoea.
This is joint work with Barbel Finkenstadt. 
________________________________________________________

Speaker: Manuele Leonelli, School of Mathematics and Statistics, University of Glasgow 

Title: Change point identification of extreme regimes 


Abstract: Precise knowledge of the tail behaviour of a distribution as well as predicting capabilities about the occurrence of extremes are fundamental in many areas of applications, and in particular for financial time series. Standard inferential routines for extremes require the imposition of arbitrary assumptions which may negatively affect the statistical estimates. The model class of extreme value mixture models, on the other hand, allows for the precise estimation of the tail of a distribution without requiring any arbitrary assumption. Here we extend this model class to handle situations where different extreme structures may be useful to perform inference over the extremes of a time series. This is achieved by proposing a novel changepoint approach for extremes, where the changepoints are estimated via Bayesian MCMC routines.  Our approach is evaluated through a series of simulations, applied to real financial data sets and assessed against competing approaches. Evidence demonstrates that the inclusion of changepoints improves the goodness of the extreme estimation in financial applications. 

Cari saluti 
Maria Piera Rogantin

--------------------------------------------------------------------------------------------------
Maria Piera Rogantin
Dipartimento di Matematica - Universita` di Genova
via Dodecaneso, 35 - 16146 GENOVA (Italia)
tel +39 010 353 6944  mob +39 3356631242   fax +39 010 353 6960
www.dima.unige.it\~rogantin
---------------------------------------------------------------------------------------------------