I am looking for formulas to calculate Kendall's tau ($\tau_\mathrm{C}$) from copula parameter (or vice versa). I am interested in extreme-t and Hüsler-Reiss copulas. It is not an issue if numerical integration is required, but I am expecting something simpler than the general formula:

$$ {\tau _{\rm{C}}} = 4 \cdot {\rm{E}}\left[ {C\left( {U,V} \right)} \right] - 1 = 4 \cdot \int\limits_0^1 {\int\limits_0^1 {C\left( {u,v} \right) \cdot {\rm{d}}C\left( {u,v} \right) - 1} } $$

$$ {\tau _{\rm{C}}} = 1 - 4 \cdot \int\limits_0^1 {\int\limits_0^1 {\frac{\partial }{{\partial u}}C\left( {u,v} \right) \cdot \frac{\partial }{{\partial v}}C\left( {u,v} \right)} } \cdot {\rm{d}}u \cdot {\rm{d}}v $$

As for Archimedean copulas the calculation can be simplified to a one-dimensional integral by using their generator function (see e.g. for Clayton copula), I hope for something similar for extreme copulas as well.

I've found the copula package that implemented extreme-t and Hüsler-Reiss copulas and functions to calculate Kendall's tau from parameter (tau()) and vice versa (iTau()). Unfortunately I was not able to extract from the source code how it calculates $\tau_\mathrm{C}$ from the copula parameter. I suspect that it interpolates between pre-computed values. I need the connection between $\tau_\mathrm{C}$ and copula parameter for $\tau_\mathrm{C}$ close to $1.0$ ($>0.98$). In this region the iTau() function of copula package gives infinite value for the parameter, which I suspect to be incorrect.



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.