Serving the Quantitative Finance Community
copula correlation and kendall tau
when we tie marginal distributions for a copula the rank correlation like kendall tau is used as a proxy and modifies suitably depending on the choice of copula.The thing that puzzles me is that the rank correlation is a single number at every point of time and comes from the residuals(feel free to correct me here and tell how exactly a rank correlation is computed in this method) at any point of time and hence the dependence,how are we actually capturing the dynamic correlation in the copula through time?
copula correlation and kendall tau
Kendalls Tau Transform is a robust estimator for the correlation coeficient. It is non-parametric and will not be influenced by outliers that much. I have seen the method where they use KTT as a point estimate for correlation and then do a ML search over the DoF in a t-copula. I think it is statistically more efficient to do an ML search over both parameters simultaniously.