Modelling Life Expectancy Gender Gap in a Multi-population Framework

This paper aims at investigating whether the life expectancy gender gap follows any long-run common tendency across different countries through a model-based analysis. If these tendencies are found to exist, then a model which takes them into account should perform better than a basic and unrestricted one. Once the gap is modeled as a multivariate non-stationary stochastic process, the goal is to find any long-run equilibrium among single series via cointegration analysis, which ultimately allows estimating some stationary linear combinations of non-stationary variables referred to as the error correction terms. To achieve such a result it is preferable to work with homogeneous samples. Therefore, the first step of this analysis consists in partitioning the initial data set into five clusters. Since the input data set includes countries with different gender gap dynamics, this diversity is clearly reflected by the difference among the models employed to fit single clusters. All series result to be non-stationary. Given the model, we check the stationarity of the error correction term and apply simple backtesting to ten-years forecasting. Evidence suggests that the fifth cluster is a cointegrated series leading to postulate that an underlying long period equilibrium does exist for this cluster.