曲線配飾方法在各國死亡率曲線預測的應用

A Curve-Fitting Approach to Modeling and Projecting Global Mortality Rates

本文研究曲線配適模型在死亡率模型上的應用。我們利用曲線配適模型來減少死亡率模型參數的數目，使得在同樣的參數數目下可以更好的估計與預測死亡率曲線。在實證上我們使用23個國家的死亡率資料，與文獻上常見的Lee-Carter模型比較預測的準確性。我們所提出的模型在男性死亡率預測上在96%的比較中有較好的成績，女性則是88%；平均的RMSE改善率分別是14.8%與8.4%。Panel Diebold-Mariano統計檢定同樣顯示我們的模型具有較佳的預測能力。最後我們將死亡率曲線的預測應用在年金商品的定價上，並說明其對年金價格的影響程度。

We introduce a curve-fitting model to reduce the dimension of age parameters and allow additional degrees of freedom to be used for the period effect. Large-scale comparisons are carried out against the benchmark Lee-Carter model. The results reveal robust improvements in forecasting accuracy across 23 sampled countries. Our approach is superior in 96 percent of male cases and 88 percent of female cases, with an average RMSE improvement of 14.8 percent and 8.4 percent respectively. The panel Diebold-Mariano test indicates that these improvements are significant. The application in insurance pricing suggests the forecasting model has significant impact on annuity price.