turkmath.org

10 years of claims data collected by Turkish Agricultural Insurance Pool (TARSIM) is analyzed to build a portfolio risk management model for internal reporting, reinsurance pricing and actuarial pricing.In order to reduce the dimension of data, product specific clusters, so called hazard regions, for hail, frost, flood, fire and storm are formed. Layer based spherical clustering methods where the layers are composed of location, number of policies issued, damage ratio and damage probability are used employing distance based spherical k-means, density based finite mixture von-Mises Fisher. Information obtained from clusters is used to build the loss distribution associated with entire portfolio. Loss distribution is estimated by Poisson-Gamma actuarial portfolio risk model to compute the economic capital, risk contributions, expected maximum loss and probable maximum loss.