Modeling multivariate operational losses via copula-based distributions with g-and-h marginals

We propose a family of copula-based multivariate distributions with g-and-h marginals. After studying the properties distribution, we develop two-step estimation strategy and analyze via simulation sampling distribution estimators. The methodology is used for analysis seven-dimensional data set containing 40 871 operational losses. empirical evidence suggests that based on single copula not flexible enough, thus model dependence structure by means vine copulas. show approach regular vines improves fit. Moreover, even though losses corresponding to different event types are found be dependent, assumption perfect positive supported our analysis. As result, value-at-risk total loss obtained from technique substantially smaller at high confidence levels respect one using common practice summing univariate value-at-risks.