EXTENDING THE LEE–CARTER MODEL WITH VARIATIONAL AUTOENCODER: A FUSION OF NEURAL NETWORK AND BAYESIAN APPROACH

Abstract In this study, we propose a nonlinear Bayesian extension of the Lee–Carter (LC) model using single-stage procedure with dimensionality reduction neural network (NN). LC is originally estimated two-stage procedure: data by singular value decomposition followed time series fitting. To address limitations LC, which are attributed to estimation and insufficient fitness data, procedures state-space (BSS) approaches extensions flexibility in modeling NNs have been proposed. As fusion these two approaches, NN variational autoencoder that performs autoencoding. Despite being parameters, our has excellent interpretability ability forecast confidence intervals, as BSS models, without Markov chain Monte Carlo methods.