Impact of different distributional assumptions in forecasting Italian mortality rates

In this paper we value the impact of different distributional assumptions relative to Lee-Carter innovations in forecasting age-specific mortality in Italy. We fit the matrix with Italy death rates from 1960 to 2004, and we observe that the innovation series presents significant kurtosis. We implement the model approximating the innovations with a symmetric Normal Inverse Gaussian (NIG) distribution for different groups of ages. We value the impact of Gaussian and NIG approximations on the distributional hypotheses considering an ex post analysis of the distributional approximation. We observe that for some age groups the NIG distributional assumption on the residuals of the LeeCarter model produces dominant results compared to the Gaussian one.