Modelling Period Effect of Lee-Carter Mortality Model with SETAR Model

Abstract

Mortality dynamics plays an important role in understanding mortality and life expectancy that will impact on economy of the countries. Many studies have considered Lee Carter method with time index as an indicator to do the forecasting. In order to forecast the mortality rates, the period index in the Lee Carter model is applied to the random walk with drift model. Despite its performance on the forecasting ability on the Lee Carter model, it is lack in term of time varying parameter that leads to higher error when fitted with random walk of drift and less accurate when forecasting the model. This is because the random walk with drift model is only adequate to data with linear series. In this study, we used the concept of non-linear time series model and then proposed a self-exciting threshold autoregressive (SETAR) to the period index. It shows that our model outperformed the random walk with drift (RWD) model for forecasting accuracy when Malaysian mortality data from 1980-2010 are considered. Long term forecasting analysis up to 2017 comparing the two models are then performed.

Keywords: drift model, Lee Carter model, random walk with drift, self-exciting threshold autoregressive.