Forecasting pension release using a fund specific experience mortality model

Mortality models that forecast deaths well can perform bad when predicting pension release. In this thesis we create two models that forecast pension release and one model that forecasts deaths. The three models are based on differences in mortality between income classes. All our data is from Dutch origin. We then develop two new backtesting methods to evaluate our models on forecasting pension release. We apply our model and on real fund data. To backtests quantify the difference between the models. Our results imply that there is no significant difference between our three models. After doing a robustness check we conclude that our results are robust.