Till innehåll på sidan

Holger Drees: Analysis of the Extremal Dependence Structure of Time Series with Applications to Financial Modeling

Tid: Må 2017-10-02 kl 15.15 - 16.15

Plats: Seminarierummet F11, Institutionen för matematik, KTH, Lindstedtsvägen 22

Medverkande: Holger Drees

Exportera till kalender

Abstract: The extremal dependence structure of a regularly varying stationary time series (Xt)t2Z is captured by the so-called spectral tail process, whose distribution is de ned via the limit of the conditional distribution of \((X_-s/|X_0|,..., X_t/|X_0|)\) given that \(|X_0| > u\). A natural estimator of the marginal cdf's of the spectral tail process can be constructed by replacing these unknown probabilities with empirical counterparts. However, Basrak and Segers (2009) established the so-called time change formula which describes the relationship between the distribution of the spectral tail process shifted in time and the distribution of the original spectral tail process. This formula can be used to derive alternative estimators for the marginal cdf's which are often more efficient than the direct estimator. The limit distributions of these estimators are too complex to directly enable the construction of confidence regions. Time permitting, we will discuss how multiplier block bootstrap can be used for this purpose. The methodology will be applied to select between GARCH-type models with distinct extreme value behavior.
(The talk is based on joint work with Richard Davis, Johan Segers and Michal Warchol)
References 
Basrak, B., and Segers, J. (2009). Regularly varying multivariate time series. Stoch. Proc. Appl. 119, 1055-1080.​