Skip to Main content Skip to Navigation

Modèles à changements de régime, applications aux données financières

Abstract : The main goal of this document is to find a good model for time series which are subject to changes in regime. The real-life application motivating this question was to have a characterization of financial crises using an Index of Market Shocks which is inspired by geophysics and some regime-switching hybrid models integrating multilayer perceptrons. The results obtained on the training data-set provided an interesting two-state separation corresponding to two different behaviours of the market, but on this occasion the quite natural questions of model selection and of choosing the "true" number of regimes also rose.
We propose to study these questions by two methods. The first one consists in proving that a penalized likelihood criterion is weakly convergent under some hypothesis on the stationarity and weak dependece of the process. The additional assumptions concerning the bracketing entropy of the class of generalized score functions are verified then in a linear Gaussian case. The second approach, rather empirical, is related to unsupervised classification methods. It combines Kohonen maps and an hierarchical clustering algorithm for which we define a new distance based on the residual sum of squares.
Document type :
Complete list of metadata
Contributor : Madalina Olteanu Connect in order to contact the contributor
Submitted on : Friday, February 23, 2007 - 4:27:56 PM
Last modification on : Wednesday, August 31, 2022 - 2:52:37 PM
Long-term archiving on: : Tuesday, April 6, 2010 - 10:13:29 PM


  • HAL Id : tel-00133132, version 1



Madalina Olteanu. Modèles à changements de régime, applications aux données financières. Mathématiques [math]. Université Panthéon-Sorbonne - Paris I, 2006. Français. ⟨tel-00133132⟩



Record views


Files downloads