Mortality Forecasting as a Flow Field in Tucker Decomposition Space

Authors: Samuel J. Clark Date: 2026-03-25 Paper ID: arxiv:2603.24299

Summary

This paper proposes a novel method for mortality forecasting that reframes the problem as integrating a flow field through the latent score space derived from a Tucker tensor decomposition of multi-population mortality data. By reducing the dimensionality of the mortality structure, the authors find that the transition primarily follows a one-dimensional flow characterized by a scalar speed function advancing the level. The system incorporates an era-weighted speed function for contemporary adaptation and uses calibrated convergence rates to guide country-specific trajectories toward a canonical structure. Evaluation via 50-year, leave-country-out cross-validation shows the approach competes effectively with established methods like Lee-Carter and Hyndman-Ullah.

Key Contributions

• Reframing mortality forecasting as integrating a flow field through the low-dimensional score space of a Tucker tensor decomposition of multi-population mortality data.
• Demonstrating that PCA reduction of the effective core matrices reveals the mortality transition is fundamentally a one-dimensional flow governed by a scalar speed function.
• Introducing an era-weighted speed function to adapt to contemporary dynamics and calibrated convergence rates to control relaxation toward a canonical mortality structure.
• Achieving competitive or superior performance against Lee-Carter and Hyndman-Ullah benchmarks over a 50-year forecast horizon in leave-country-out cross-validation.

Limitations

The reliance on the structure captured by the Tucker decomposition might limit the ability to model entirely novel, abrupt shifts in mortality dynamics that fall outside the learned low-dimensional space.

Open Questions & Future Work

• mortality-trajectory-very-low-levels
• hybrid-short-long-horizon-forecasting
• generalizability-beyond-european-hmd
• data-scarcity-super-low-mortality
• adaptive-era-weighting-kernels
• conditioning-on-covariates
• fully-bayesian-uncertainty-quantification
• era-weighting-extensions
• hybrid-short-horizon-system
• modelife-table-byproduct-use

Key Concepts

• Tucker Decomposition for Mortality Forecasting: A method reframing mortality forecasting as integrating a flow field within the low-dimensional score space derived from a Tucker tensor decomposition of multi-population mortality data.

Datasets

• Human Mortality Database

Limitations

The reliance on the structure captured by the Tucker decomposition might limit the ability to model entirely novel, abrupt shifts in mortality dynamics that fall outside the learned low-dimensional space.

Links

• ArXiv Abstract
• PDF