Kernel Independent Component Analysis (ICA) for identification of structural innovations in multivariate GARCH models

Details

Supervisors

Hafner, Christian

Faculty

Faculté des sciences

Degree label

Master [120] en science des données, orientation statistique, à finalité spécialisée

Abstract

Structural models reflect the contemporaneous and dynamic links between a set of time series, allowing e.g. to address inference of causality and impulse response functions. For identification, suitable parameter constraints need to be imposed. Theory-based constraints such as sign restrictions are common, especially in econometric applications. Recent research has concentrated on data driven identification constraints, such as independent component analysis (ICA). Thus, in this thesis, we investigate potential benefits of using nonlinear ICA such as kernel ICA of Bash & Jordan (2002) for identification. It contains a review of structural shocks, ICA and kernel ICA. A simulation study is performed to evaluate the performance of nonlinear ICA over linear ICA in structural shocks identification. And finally, an application on financial indexes.