Skew mixture models for loss distributions: A Bayesian approach