Hierarchical forecasting of Italian mortality data with gender and cause of death reconciliation

We introduce a novel approach that integrates Compositional Data Analysis (CoDA) with forecast reconciliation to improve mortality projections disaggregated by cause of death and demographic groups. By leveraging the hierarchical and grouped structures inherent in causes-of-death mortality data, our framework addresses key challenges in mortality forecasting, ensuring internal coherence across different levels of aggregation while maintaining a high degree of predictive accuracy. Our empirical results on Italian cause-specific mortality data, extracted from the WHO Database, demonstrate that forecast reconciliation significantly enhances the reliability of projections. Furthermore, the MinT approach consistently outperforms alternative methods, particularly for top-level mortality forecasts. Our study highlights the substantial benefits of combining CoDA with forecast reconciliation, with broad implications for demographic research, public health planning, and actuarial assessments.