FORECASTING STOCHASTIC MORTALITY RATE AND LONGEVITY RISK MEASURE

• Main Author: HIDAYAT KALLA (NIM : 20814024), YUSUF
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/24835
• Institution: Institut Teknologi Bandung

Life insurance company and pension funds require predictive mortality rate to calculate premiums and claims reserves. Meanwhile, realized mortality rate determine claim expense of company. The difference value between realized and expected mortality rates leads to potential losses to insurer. Therefore, mortality rate must be modeled into a good forecasting model. One model that is often used is AR(1)-GARCH(1,1) stochastic model. In addition to the use good model, parameter estimation method must also be considered. A appropriate parameter estimation method is a method that unbiased and mean square error is relatively small.In this thesis, Quasi-Maximum Likelihood(QML) method is carried out. In addition to the be a appropriate estimation method, this method is used because it provides greater exibility in the error distribution assumption. Adopting mortality data in Australian form 1950-2011, predictions are carried out with using multistep-forecast and the accuracy test is Conditional Mean Square Error of Prediction (CMSPE). Results of this thesis show that AR(1)-GARCH(1,1) model is a good mortality model for modeling mortality rate of Australian. Meanwhile, predictive mortality rate by using QML method have mean square error smaller than other. Thus, AR(1)-GARCH(1,1) model by using QML method is an accurate model for forecasting mortality rate. Furthermore, this model is using to measure the longevity risk of Australian. This risk relates to the potential losses the company on annuity product life. The risk measures have been exploited such as Value-at-Risk (VaR) and Conditional Var (CVaR). VaR and CVaR are minimum claims reserves that company should provide to minimize the longevity risk.