Quantification of the estimation risk inherent in loss distribution approach models