A comparison of tail dependence estimators

We review several commonly used methods for estimating the tail dependence in a given data sample. In simulations, we show that especially static estimators produce severely biased estimates of tail dependence when applied to samples with time-varying extreme dependence. In some instances, using static estimators for time-varying data leads to estimates more than twice as high as the true tail dependence. Our findings attenuate the need to account for the time-variation in extreme dependence by using dynamic models. Taking all simulations into account, the dynamic tail dependence estimators perform best with the Dynamic Symmetric Copula (DSC) taking the lead. We test our findings in an empirical study and show that the choice of estimator significantly affects the importance of tail dependence for asset prices.