Clustering of mortality paths with the Hellinger distance and visualization through the DISTATIS technique

Stochastic mortality models improve forecast accuracy through multipopulation approaches, yet lack rigorous criteria for country selection. This study introduces a novel, distance-based method using Hellinger distance and hierarchical clustering to identify countries with similar average mortality. Convergence or divergence of mortality paths is then checked visually by projecting the distances between countries in different years to a common Cartesian space using the DISTATIS technique. Analyzing mortality data from 1960 to 2019 for multiple countries from the Human Mortality Database via hierarchical clustering and DISTATIS visualization, I identify stable clusters and reveal convergence trends that are subsequently described through mortality indicators. The Hellinger distance outperforms other plausible choices of distances and the DISTATIS factors capture both the timing and dispersion of mortality. The findings offer a robust measure for country selection in multipopulation models, improving on the evaluation of convergence or divergence of mortality paths compared to methods based on life expectancy.