Local and global mortality experience: A novel hierarchical model for regional mortality risk

Abstract

Accurate mortality risk assessment is critical for decision-making in life insurance, healthcare, and public policy. Regional variability in mortality, driven by diverse local factors and inconsistent data availability, presents significant modeling challenges. This study introduces a novel hierarchical mortality risk model that integrates global and local data, enhancing regional mortality estimation across diverse regions. The proposed approach employs a two-stage process: first, a global Light Gradient Boosting Machine model is trained on globally shared features; second, region-specific models are developed to incorporate local characteristics. This framework outperforms both purely local models and standard imputation techniques, particularly in data-scarce regions, by leveraging global patterns to improve generalization. The model is computationally efficient, scalable, and robust in handling missing values, making it adaptable for other domains requiring integration of multi-regional data. This method enhances predictive accuracy across various regions and provides a more reliable approach for mortality risk estimation in data-scarce environments.