The purpose of this master's thesis is to discuss statistical solutions to the following problem: Given a multidimensional data sample, how can the associated copula be identified? In particular the focus lies on methods which need none or only few prior assumptions. Non-parametric estimation procedures and hypothesis tests for selected copula properties are presented. The discussion covers hypothesis tests for Archimedeanity, Exchangeability and Radial Symmetry, as well as estimators of concordance measures and tail dependence coefficients. Furthermore, parameter estimators for parametric copula families and goodness-of-fit tests are considered. A simulation study analyzes the finite-sample performance of the presented methods. The thesis starts with the discussion of the theoretical background. An introduction to the theory of copulas is given and the empirical copula, a non-parametric estimator, is defined. Furthermore, its asymptotic properties are derived using the theory of weak convergence for empirical processes. The following chapters introduce the statistical hypothesis tests and estimation procedures. The test statistics and estimators are defined, their asymptotic properties are derived and multiplier bootstrap techniques are discussed, allowing to handle their complex asymptotic distributions. The results of the simulation study are presented in the final chapter. They suggest that the presented methods provide helpful tools if the dependence structure of a random sample has to be analyzed. The finite-sample performance of the hypothesis tests and estimation procedures is found to be convincing. One exception is the estimation of tail dependence coefficients. This estimator should be used with care and a preceding test for tail independence is advisable.
«
The purpose of this master's thesis is to discuss statistical solutions to the following problem: Given a multidimensional data sample, how can the associated copula be identified? In particular the focus lies on methods which need none or only few prior assumptions. Non-parametric estimation procedures and hypothesis tests for selected copula properties are presented. The discussion covers hypothesis tests for Archimedeanity, Exchangeability and Radial Symmetry, as well as estimators of concord...
»