A Comparison Of Stochastic Mortality Models Using Chinese Data

Traditionally, actuaries utilize life table to price life insurance products. The practice assumes mortality is static, which means that it is not changing over time. However, according to data from many countries, mortality had improved greatly during the20th century, and it is clearly a stochastic process. During the last two decades, many stochastic mortality models have been introduced. To price mortality related products, it is crucial for us to find the most suitable stochastic mortality model among numerous existing models for the Chinese old population we concerns.In the article, I compare qualitatively and quantitatively eight representative stochastic mortality models concluding Lee-Carter model, Renshaw-Haberman model, Currie model, Carins-Blake-Dawd model and3extensions of it and a high age version of Plat model in high age mortality rates of China. The comparison involves three aspects-1) properties of the structure of models (parsimony, ease of implementation, correlation term structure, etc.),2)goodness of fit (BIC/AIC/HQ, likelihood ratio test, standard residual normality, robustness of parameters), and3)goodness of forecast (plausibility of forecasts, robustness of forecasts, out-of-sample test).The original CBD model is the only model performs well under all comparisons or tests in the high age male Chinese dataset, and hence is the most suitable model. In last chapter, I try introducing a new model intending to overperform CBD model in the dataset concerned, and only to find it underperforms the CBD model.