Yechao Meng
University of Prince Edward Island
Title
Mortality Prediction via Age-Specific Band Selection
Abstract
Longevity risk, driven by increasing life expectancy, presents a major challenge for insurers, governments, and individuals. Inaccurate mortality projections can lead to pricing and valuation errors in life insurance and living benefits products, which can strain pension funds and annuity providers. An analysis of mortality data from the Human Mortality Database highlights similar patterns of development within certain age groups (e.g., young, middle-aged, and senior) and differences across them. Recent research suggests that these similarities in age-specific mortality trends can improve the accuracy of mortality models. However, traditional approaches in mortality literature model all ages together, attempting to capture these patterns by adding more parameters. This would overlook the potential drawbacks of using data with an inappropriate scope and incorporating data with conflicting signals, thus weakening the model's performance.
Instead of increasing model complexity to account for these age-specific patterns, we propose a different approach that focuses on selecting the most relevant age ranges. This preserves model simplicity while enhancing effectiveness. The innovation lies in an age-specific solution, where an age-specific band is used to borrow information from "neighboring" ages and build prediction models for each individual age in a mortality table. By carefully screening data to balance relevant signals and exclude noise, we improve the accuracy of mortality rate predictions. This concept is further extended to share information across multiple populations and ages simultaneously. Extensive numerical analyses using the Human Mortality Database (HMD) consistently show improvements in prediction accuracy across various scenarios.
