Fitting insurance claims to skewed distributions : Are the skew-normal and skew-student good models?