A Multi-population Locally-Coherent Mortality Model

This paper proposes a simple and fully-interpretable neural network model for multi-population mortality modelling and forecasting. The architecture is designed to be interpretable in the Lee-Carter framework and induces a massive reduction of the parameters to optimise. The model structure leads the creation of clusters of countries with similar mortality patterns during the fitting procedure highlighting differences and commonalities among the clusters. Numerical experiments performed on the Human Mortality Database Data show that the proposed model produces reliable estimates and very accurate forecasts.