Spatial Prediction of Mortality Based on Double-layer Gaussian Process Ensemble Regression Model

Abstract

 Mortality rate prediction is one of the core aspects of risk pricing and product design in the insurance industry. Its accuracy directly determines the rationality of premiums, the adequacy of reserves, and the solvency assessment for life insurance, annuity products and other personal insurance products. Currently, there is temporal and spatial heterogeneity in China's population mortality rate, and implementing a traditional unified rate system is difficult to match the actual risk distribution, which may lead to systematic pricing deviations and other problems. This paper proposes an ensemble learning model based on Gaussian process regression, integrating regional factors as feature variables. On the one hand, it uses ensemble learning methods to further optimize the prediction results of the Gaussian process regression model; on the other hand, it uses Gaussian process regression to quantitatively estimate the uncertainty of future predictions. In the actual data analysis, this paper first verified that there are differences in population mortality rates among 31 provinces in China; secondly, it conducted a Monte Carlo experiment on the proposed method and found that this method can fit nonlinear functions; then, based on the historical mortality rates of 31 provinces in China, it conducted real data analysis and prediction to obtain mortality rate prediction intervals for uncertainty quantification estimation. From the prediction results, this method can well support spatial prediction of population mortality rates, thereby providing more effective decision-making basis for insurance companies to set premiums for personal insurance products.