Lee-Carter Model and Extensions to Forecast Thai Mortality Rate

Abstract

This paper aims to forecast Thai mortality rate using the Lee-Carter and two extensions model: age-period-cohort model and specific case of age-period-cohort model. The parameters of models are estimated by by minimizing the negative binomial deviance to fit the parameters. The assumptions of the death count based on the equal-dispersion and the over-dispersion, a Poisson and Negative binomial distribution respectively. The Score Statistics are used to detect over-dispersion. The iterative fitting model optimized by the Poisson maximum likelihood and Negative binomial maximum likelihood by considering the associated deviance. We fit the models into Thailand population. The appropriate model is selected by Mean Absolute Error (MAE), Mean Square Error (MSE), Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE). The Box-Jenkins methodology was applied to estimate and forecast time series index with the appropriate ARIMA time series model. The study found that Thai mortality rate trend to decrease according to increasing year both male and female.