We introduce a new methology for modelling financial durations of ultra-high frequency data using copulas.
While the class of common ACD models are chiefly characterized by strict parameterizations and computational burdens, the semiparametric copula approach proposed
here offers more flexibility in modelling the dynamic duration process by separating the marginal distributions of waiting times from their temporal dependence structure.
Comparing both frameworks with density forecast abilities, the SCoD model clearly shows a better performance.