Major insurance claims may take many years to settle, making it challenging for insurers to use recent, relevant data in the estimation of their paid amounts model. Including only closed files in inference creates sampling bias towards simple cases, and using all files but ignoring claim status also understates the risk. We propose a Bayesian model for individual claims and a multiple imputation procedure for exploiting the information in censored claims. The dependence between claimants involved in the same accident and between coverages is modeled with a copula. The approach is illustrated with data from a Canadian insurance company.
Coffee and biscuits will be offered at the beginning of the seminar.
Welcome to everyone!