Phd in SCHOOL OF STATISTICAL SCIENCES

Thesis title: Mortality neural forecasting

Predicting mortality is a major challenge for both demographers and actuaries. The latter need to anticipate various future mortality scenarios with the greatest possible accuracy, as in the case of annuities pricing and longevity risk assessments. However, the current wide range of stochastic mortality models highlights some deficiencies in predicting future mortality realizations, particularly when accelerations or decelerations of longevity occur. The aim of this research thesis is to investigate the adequacy of a new mortality forecasting approach based on artificial Neural Networks. To this end, after an examination of the theoretical Neural Networks fundamentals, the present work shows the Neural Networks competitiveness in predicting the future dynamics of human mortality, also allowing the efficacy of already existing predictive models, such as the canonical Lee-Carter model. Therefore, our data-driven proposal contributes to the mortality literature as new advance in mortality forecasting, that is the neural forecasting approach.