This paper introduces a nonparametric estimator for tail dependence in the constant conditional correlation GARCH framework, in contrast to existing estimators that impose the iid assumption. So long as stationarity is satisfied, the difference between the distribution of the tail dependence estimator under the iid and GARCH case is a scaling variance. Without the scaling variance, tests based on this estimator overreject the null of asymptotic tail dependence. An empirical application to tail dependence between emerging market bonds and equities shows that there is tail depenedence in their joint density even though the standard linear correlation coefficients indicate low correlation between assets. These findings and the methods introduced here have implications for risk management and portfolio allocation theory that are based on the standard correlation estimato
