Change-point detection for financial time series models
Time and place
1 PM on Tuesday, March 31st, 2009; NAC 1/511E
Dr. В Siegfried Hoermann (University of Utah, Salt Lake City)
Abstract
Many standard methods in time series analysis work
under the assumption that the data investigated have been sampled from some weakly
stationary process. Especially, this is a crucial requirement in order that we can
produce meaningful estimates and reliable forecasts. It is therefore of considerable interest
to detect structural breaks in the data before any other statistical analysis is carried
out. In order to apply common procedures of change-point analysis we first have to understand
the dependence and dynamics of the data generating processes. The purpose of this talk
is to have a closer look at the structure of several financial time series models and to show
how these results can be used to identify breaks in the volatilities of univariate and
cross-volatilities of multivariate sequences. We propose a non-parametric method and show that our
results are widely applicable, e.g. for augmented GARCH sequences,
multivariate ARMA and dierent multivariate GARCH models
