Student t copula
Models multivariate dependence with joint upper- and lower-tail dependence.
How it works
A Student t copula combines marginal distributions with dependence from a multivariate t distribution. Its correlation parameters govern co-movement in the latent variables, while common scale variation permits joint extremes. With finite degrees of freedom and non-degenerate correlation, the standard t copula has symmetric upper and lower tail dependence.
Worked example
Fit a Gaussian and a t copula to the same rank dependence. Simulate two obligors and count the fraction of cases where the second is also below its first percentile given that the first is below that threshold. The model with common tail dependence can produce more joint extreme observations without changing either marginal.
Validation and limitations
Validate rank dependence and tail co-exceedances separately. The degrees of freedom can be weakly identified in short histories. Symmetric tail dependence may be unsuitable when downside extremes behave differently from upside extremes. Stress the parameters before using them for credit or portfolio tail decisions.