Statistical features of persistence and long memory in mortality data

Abstract Understanding core statistical properties and data features in mortality are fundamental to the development of machine learning methods for demographic actuarial applications projection. The study such forms basis classification, regression forecasting tasks. In particular, understanding key structure can aid improving accuracy undertaking projection when constructing life tables. ability accurately forecast is a critical aspect demography, insurance product design pricing, pension planning insurance-based decision risk management. Though many stylised facts have been discussed literature, we provide evidence novel feature that pervasive at national level as yet unexplored. this regard, demonstrate work strong existence long memory data, second structures display multifractality act discriminator dynamics by age, gender country. To achieve this, first outline way which choose represent persistence from an estimator perspective. We make natural link between class attribute stochastic processes based on fractional Brownian motion. This allows us use well established estimators Hurst exponent then robustly data. introduce analysis notion science known multifractality. different timescales. its sample sizes commensurate with national-level age term historical records. A series synthetic studies comprehensive real death count studied order pervasiveness both mono-fractal multifractal functional verified be present most countries groups gender. conclude demonstrating how used kernel clustering model improve these applications.