Integrating Genetic Algorithm and Markov Switching Regression to Construct the Efficient Frontier of Equity-Linked Life Insurance Policies

[[abstract]]本計畫結合基因演算法以及馬可夫轉換迴歸模型，企圖找尋投資型保險商品之效率前緣。本計畫探討台灣現行投資型保險投資管理辦法及證券投資信託基金管理辦法對台灣投資型保險效率前緣的影響，並討論在該規範限制下，基因演算法以及馬可夫轉換模型所配置之投資型保險商品的最適效率及其預測能力。本計畫初步發現在現行投資型保險投資管理辦法以及證券投資信託基金管理辦法規範下，台灣投資型保險商品並未達到效率前緣，而以基因演算法所建構的投資型保險商品確能創造較高報酬與較低風險。另外本計畫以馬可夫轉換迴歸找出不同風險與報酬狀態分類下，投資型保險商品投資標的之顯著報酬變數，作為基因演算法判定投資型保險商品之保單持有者決定投資資產權重大小的依據。並進一步探討以兩狀態馬可夫轉換迴歸所估計之預期報酬及變異共變異矩陣，做為基因演算法之模型目標函數的投入因素時，是否能獲得更好的績效表現。[[abstract]]This project employs Genetic Algorithm (GA）to construct the efficiency frontier for the Equity-linked life insurance based on The Administration Acts of Security Investment and Trust Fund in Taiwan. The model proposed in this paper can be used to examine the effects of the Administration Acts on efficiency frontier of different Equity-linked life insurance and to verify the profit prediction power of alternative Equity-linked life insurance constructed by GA. The preliminary findings of this project show that the current Equity-linked life insurance based on The Administration Act of Security Investment and Trust Fund in Taiwan exhibits less efficiency. Compared to the performance of Taiwan Stock Exchange Corporation (TSEC) weighted price index, the average performance of Equity-linked life insurance and stock funds within the same period, the performance of the Equity-linked life insurance constructed in this paper is more efficient across the test sample. In addition, this project employs Markov switching regression to identify variables that are of statistical significance concerning the return and variance of the investment objects of the Equity-linked life insurance under different states. Furthermore, these variables are use to order the weight of the investment objects and to construct the product design of the Equity-linked life insurance. Finally, this project shows that the Equity-linked life insurance combined with the predicted return and variance-covariance matrix of Markov switching regression will offer a better and stable prediction power.[[note]]NSC98-2410-H327-02