Performance evaluation of the multi-dimensional Bühlmann credibility approach in predicting multi-population mortality rates

Abstract

Mortality prediction is a crucial aspect for insurance and pension fund companies in deciding a suitable premium. The aim of this research is to discuss a cross-country (multi-population) mortality modeling in order to obtain a better mortality prediction. This modeling is based on the multi-dimensional Buhlmann credibility approach. The expansion in this research refers to mortality rate data taken from several countries. The Buhlmann credibility theory is generally used to predict the value of a random variable in a given period in the future. In this research, prediction for years to come was done using two strategies: Expanding Window and Moving Window. For every prediction in the upcoming period, both Expanding Window and Moving Window use prediction result values as additional data to build upon the prediction model for the next year; however, Moving Window also dismisses the oldest data. The model parameter is estimated with non-parametric approach. This model is then applied to the mortality data from Japan, Sweden, and the Czech Republic. Finally, each model’s performance is analyzed using Mean Absolute Percentage Error (MAPE) and Average Mean Absolute Percentage Error (AMAPE). The result shows that the performance of the multi-dimensional Buhlmann credibility approach is satisfactory in modeling cross-country mortality rates.