This thesis deals with various aspects of multivariate dependencies and extreme values. First, a new dependence measure is developed and non-parametrical estimator as well as a concept of visualization of extremes. Also, tail-copula estimators are developed under the assumptions of having either elliptical distributions or elliptical copula. Of these, the asymptotic behavior of first and second order is determined and the improvements compared to non-parametric estimators are shown both theoretical and by simulation. Another chapter extends correlation structure analysis to copulae. Therefore, copula-based estimators are designed and their asymptotic behavior is shown, where a tail-copula estimator allows for a structure analysis of extremes. Finally, the extreme value distribution of a credit default portfolio is proven for a wide and flexible class of default distributions and an improved concept of fitting the portfolio-distribution is shown.
«
This thesis deals with various aspects of multivariate dependencies and extreme values. First, a new dependence measure is developed and non-parametrical estimator as well as a concept of visualization of extremes. Also, tail-copula estimators are developed under the assumptions of having either elliptical distributions or elliptical copula. Of these, the asymptotic behavior of first and second order is determined and the improvements compared to non-parametric estimators are shown both theoreti...
»