Forecasting Malaysian mortality rates using the Lee-Carter model with fitting period variants

Abstract

This study aims to forecast Malaysian mortality rates by age and gender using the well-known Lee-Carter model. Data obtained from the Department of Statistics Malaysia which consists of central mortality rates by age and gender from year 1970 to 2018. Two different sets of fitting periods were determined based on the observations of changes in mortality index patterns over the years. The Set A consists of 24-year fitting period from 1970 to 1993 whereas Set B comprises 31-year fitting period from 1970 to 2000. The in-sample evaluation of the Lee-Carter model are performed using the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) whereas the out-sample errors are calculated using the Mean Absolute Percent Errors (MAPE) and Root Mean Square Error (RMSE). Results show that the set A is better fitted into the Lee-Carter model than that of set B by having lower values of AIC and BIC, consequently produced more accurate out-sample forecast values for females. The Lee-Carter model is a reliable model for Malaysia data, however careful attention must be given when selecting the best fitting period.