Explaining Regional Mortality Differences with an Economic-Neural Model: Evidence from European NUTS-2 Regions

This article introduces a novel framework for explaining regional mortality differences across European NUTS-2 areas using macroeconomic indicators. Because regional death rates are substantially noisier than national aggregates, we model age-specific mortality as a smooth B-spline surface whose coefficients are predicted by a feed-forward neural network. The network takes as inputs a set of interpretable regional factors, including GDP per capita, purchasing power, employment rate, educational attainment, and NO2 emissions. Model parameters are estimated via maximization of the Poisson log-likelihood, and the methodology is applied to French regional mortality data. Compared with the LiLee multi-population framework, the proposed approach offers several advantages. First, it provides an interpretable link between economic conditions and mortality, allowing the impact of policy-relevant variables to be quantied. Second, the combination of neural networks with B-splines yields smooth, stable mortality curves and avoids the overtting often observed in non-parametric regional models. Finally, the model is suciently robust for long-term mortality forecasting and actuarial applications such as life expectancy projections and annuity valuation.