對於商業保險公司和政府單位而言，死亡率的改善和未來死亡率的預估一直是一大重要議題。特別是對於退休金相關的社會保險、勞退或是商業年金、壽險等等，如何找尋一個準確的預估模式對未來的死亡率改善情況進行預測，並釐訂合理的保費及提列適當的準備金，是對於一個保險制度能否永續經營的重要因素。過去所使用的配適方法，大多僅以單一群體的過去資料輔助未來的預測，例如 Li and Carter (1992)所提出的 Lee-Carter Model，或是 Bell (1997)使用主成分分析法 (Principal Component Analysis, PCA)等僅針對單一群體本身變數進行分析之方式。然而綜觀全球死亡率改善趨勢，可發現國與國間、組與祖間雖有不同，但仍具備共同的趨勢。因此在考慮未來的死亡率配適方面，應加入組與組間的共同因子 (common factors) 進行考量。 Li and Lee (2005)曾提出 Augmented Lee-Carter Model，即對原本的Lee-Carter Model進行修正，加入共同因素項，並且得到更好的預測效果。 本文則採用考慮共同因子之主成分分析原理建構多重群體死亡率模型，即透過主成分分析法，同時考慮不同群體間的死亡率，並以台灣男性和女性1970年至2010年的死亡率資料，做為兩個子群體進行分析。本文使用之主成分分析法模式，和 Lee-Carter Model (Li and Carter, 1992) 和 Augmented Lee-Carter Model (Li and Lee, 2005)，以MAPE法對個別的預測能力進行分析，並得出採用PCA的模式，在預測男性短年期(5年)內的預估能力屬精確(MAPE 介於10%~20%之間)，然而在長期預估下容易失準，且所有使用的模型，在配適台灣資料時皆發生無法準確預估嬰幼兒期(0~3歲)和老年期(80歲以上)之情形。本文並以所有模型預估之死亡率計算保險公司之準備金與保費提列，並與第五回經驗生命表進行比較。 / For governments and life insurance companies, mortality rates are one of the key factors in determining premiums and reserves. Ignoring or miscalculating mortality rates might have negative influences in pricing. However, most of the mortality models do not consider the common trends between groups. In this article, we try to construct the mortality structure which considering common trends of multi-groups populations with principal component analysis (PCA) method. We choose 9 factors to set up our model and fit with the actual data in Taiwan’s gender mortality. We also compare the Lee-Carter Model (Lee and Carter, 1992) and the augmented Lee-Carter Model (Li and Hardy, 2012) with our common factors PCA model, and we find that the PCA model has the least MAPE than other model in five years forecasting in both genders. After finishing basic analysis, we use the mortality data of Taiwan (1970 to 2010) from human mortality database to construct the life expectancy model. We adopt the same criteria to choose the components we need. We also compare the level premium and reserves by different forecasting mortality rates. All of the models indicate life insurance companies to provide higher reserves and level premium than using the 5th TSO experience mortality rare. We will do following research by using company-specific data to construct unique life expectancy model.

近年隨著油價、物價上漲所導致的通貨膨脹風險，壽險業者以增額型終身壽險來吸引潛在消費者。另外，由於醫療技術的進步，使得死亡率逐年改善，因此將造成保單在設計時可能將遭受到長壽風險的影響。本篇文章的主要目的即探討長壽風險對於保單責任準備金的影響，並以增額型終身壽險作為本文主要分析標的。首先建構死亡率模型(Lee-Carter模型)，用來配適並模擬死亡率，接著探討增額型終身壽險在各保單年度下之現金流量以及責任準備金的提存，進一步再引進不同的死亡率來探討其現金流量分佈情形與責任準備金之提存。本文研究結果發現，在保險公司未採用遞迴方式計算保費時，當繳費期間愈短、複利利率愈高以及投保年齡愈低時，保險公司所面臨之長壽風險愈大，其後在帶入各種不同死亡率模型，發現死亡改善率愈高，保險公司所面臨之長壽風險愈大，而保險公司在提存責任準備金時，並未考慮到死亡改善率的部分，此對保險公司的財務健全將造成隱憂，本文於此部分建議監理機關透過法規(RBC)的制訂，調整準備金提存的係數，以降低長壽風險對保險公司財務之衝擊。 關鍵字：長壽風險、死亡率模型、增額型終身壽險、保單責任準備金、增額準備金、Lee-Carter Model以及RBC制度。 / With the improvement of medical technology, the life expectancy around the world is increasing year by year during the past decade. Therefore, the increasing whole life insurance policy is popular during these years because its benefits are escalating with time and policyholders think they could gain more benefits when they live longer. Like annuity policies, the increasing whole life insurance could also suffer from the longevity risk, which may have enormous impact on the financial statements of insurers. The purpose of this paper is to discuss the impact of longevity risk on reserves, based on increasing whole life insurance policy. First, we construct Lee-Carter model to fit and simulate mortality rate and assume different mortality improvements from the 2002 Taiwan Standard Ordinary Experience Mortality Table (2002TSO) for further comparisons. And then, we construct a simple model to analyze the cash flows of the increasing whole life policies based on the mortality rates we observed. By constructing a simple model and simulation, we find that if the insurance company does not correctly estimate longevity risk, the insurance company will lose money on the increasing whole life policies. In order to mitigate the insufficiency of life insurers for the increasing whole life policies, we try to provide some supervision suggestion from the view of the risk-based capital (RBC) requirements. We calculate the factor of insurance risk (C2) of RBC requirements because this factor represents the surplus needed to provide for excess claims over expected, both from random fluctuations and from inaccurate pricing for future levels of claims. Keywords: longevity risk, increasing whole life insurance policy, Lee-Carter model, risk-based capital (RBC).

本篇論文將厚尾分配（Heavy-Tailed Distribution）應用在財務及保險精算上。本研究主要有三個部分：第一部份是用厚尾分配來重新建構Lee-Carter模型（1992），發現改良後的Lee-Carter模型其配適與預測效果都較準確。第二部分是將厚尾分配建構於具有世代因子（Cohort Factor）的Renshaw and Haberman模型（2006）中，其配適及預測效果皆有顯著改善，此外，針對英格蘭及威爾斯（England and Wales）訂價長壽交換（Longevity Swaps），結果顯示此模型可以支付較少的長壽交換之保費以及避免低估損失準備金。第三部分是財務上的應用，利用Schmidt等人（2006）提出的多元仿射廣義雙曲線分配（Multivariate Affine Generalized Hyperbolic Distributions; MAGH）於Boyle等人（2003）提出的低偏差網狀法（Low Discrepancy Mesh; LDM）來定價多維度的百慕達選擇權。理論上，LDM法的數值會高於Longstaff and Schwartz（2001）提出的最小平方法（Least Square Method; LSM）的數值，而數值分析結果皆一致顯示此性質，藉由此特性，我們可知道多維度之百慕達選擇權的真值落於此範圍之間。 / The thesis focus on the application of heavy-tailed distributions in finance and actuarial science. We provide three applications in this thesis. The first application is that we refine the Lee-Carter model (1992) with heavy-tailed distributions. The results show that the Lee-Carter model with heavy-tailed distributions provide better fitting and prediction. The second application is that we also model the error term of Renshaw and Haberman model (2006) using heavy-tailed distributions and provide an iterative fitting algorithm to generate maximum likelihood estimates under the Cox regression model. Using the RH model with non-Gaussian innovations can pay lower premiums of longevity swaps and avoid the underestimation of loss reserves for England and Wales. The third application is that we use multivariate affine generalized hyperbolic (MAGH) distributions introduced by Schmidt et al. (2006) and low discrepancy mesh (LDM) method introduced by Boyle et al. (2003), to show how to price multidimensional Bermudan derivatives. In addition, the LDM estimates are higher than the corresponding estimates from the Least Square Method (LSM) of Longstaff and Schwartz (2001). This is consistent with the property that the LDM estimate is high bias while the LSM estimate is low bias. This property also ensures that the true option value will lie between these two bounds.