1. NCCU Academic Hub
  2. Publications
  3. College of Commerce
  4. NSC Projects
  5. Item

Publications-NSC Projects

Article View/Open

Publication Export

Google ScholarTM

NCCU Library

Citation Infomation

Related Publications in TAIR

題名 建立內部模型管理壽險公司的死亡率風險
Managing a Life Insurer’S Mortality Risk by Building Internal Models
作者 蔡政憲
貢獻者 風管系
日期 2020-01
上傳時間 17-Jun-2026 15:55:15 (UTC+8)
摘要 本計畫預定使用國內某大型壽險公司的終身壽險保單資訊，建立三套死亡率模型。此資料庫中有上千萬張保單，可看到投保人與被保險人的資訊（例如投保年齡、職業、體位、居住地區等）以及保單的資訊（例如保額、保費、繳別、佣金率、通路、附約），資料期間是1994~2014。 我們將於計畫期間的第一年運用這個資料庫探討：有哪些個人特質與商品特徵的變數會影響死亡的機率。過去的文獻多是用整體人口的樣本資料，集中於探討個人特質的影響；此計畫則是運用投保人/被保險人的資料，其所含有的個人資料與商品特質比文獻中所使用的詳盡得多。預計遵循文獻採用離散型的Cox regression估計存活機率，估計的方法將為最大概似法。 第二年的計畫則是運用此資料庫建構死亡率的推估模型。由於此資料庫的期間還不夠長，因此我們的第一步是先運用全國死亡率的資料，建構因子模型，再對所找出來的因子建構時間序列模型，即可進行全國死亡率的推估。在建構時間序列模型時，我們將根據文獻加入一些總體經濟的變數。第二步則是從資料庫中整理出各歷年度的死亡率表/曲線。第三步則是建立公司的死亡率曲線和全國曲線之間的關係。我們將採用關連性模型（Relational Modeling）以及Panel Regression來建立這兩者之間的關係。 第三年的計畫則是運用多群因素分析法於所整理出來的七類商品的死亡率曲線以及全國的死亡率曲線。預計能找出跨群的共同因子以及分屬個別群的獨特因子。透過這樣的分析，我們可以更進一步知道第二年計畫中所找出的總經變數對個別商品影響力之差異；而所找到的獨特因子將對商品間的風險衡量與管理有顯著的意義。 由於本計畫所建立的模型是使用公司的內部資料，建好的模型也將給公司使用，因此算是公司的內部模型。此公司未來也將使用這些模型來估計死亡率的風險，進而估算出死亡率的經濟資本。
We plan to build three mortality models based on the proprietary information provided by a major Taiwanese company. Currently, the database comprised around ten million contracts, including all kinds of whole life insurance contracts. This database includes detailed information of policyholder and insured such as age, occupation, residential area, and BMI, and also the information of each contract such as coverage, premium, commission, distribution, and rider with the main contract. The sampling period starts from 1994 to 2014 and still expanding. In the first year of this project, we investigate the relationship between mortality rate and insured’s personal and contractual characteristics. Past literature mostly focused on the personal aspect, with aggregate data. Our project capitalizes on the much more detailed insured data than the past literature. Following the previous literature, we use the Cox regression to find the variable that explains the differential mortality rate. We estimate the model with maximum likelihood. The variable we identified will affect the risk of premium, reserve, and contract in general. The goal of second-year project is to build a model that allows us to forecast the mortality rate. The length of dataset is not long enough to support a forecasting model on its own. Therefore, the first step is to estimate a mortality model using the national population data. Then we estimate a relational model that connects the insured population to the national population so that we can extrapolate the dynamic to insured population. The third year’s goal is to analyze insured population and general population with multi-group factor analysis. We expect to find some common factors that exist among all groups and unique factors that belong to each individual group. Our analysis will focus on the impact of variable we found in the second year project to each insurance product. The unique factor we found would provide interesting information to risk management and pricing of each product. Because the data we used is provided by the life insurance company, we will provide our model to the company as an internal model. This is why they are willing to provide us the data. They will use these models to estimate the mortality risk and economic capital.
關聯 科技部, MOST105-2410-H004-070-MY3, 105.08-108.07
資料類型 report
dc.contributor 風管系
dc.creator (作者) 蔡政憲
dc.date (日期) 2020-01
dc.date.accessioned 17-Jun-2026 15:55:15 (UTC+8)-
dc.date.available 17-Jun-2026 15:55:15 (UTC+8)-
dc.date.issued (上傳時間) 17-Jun-2026 15:55:15 (UTC+8)-
dc.identifier.uri (URI) https://ah.lib.nccu.edu.tw/item?item_id=182989-
dc.description.abstract (摘要) 本計畫預定使用國內某大型壽險公司的終身壽險保單資訊，建立三套死亡率模型。此資料庫中有上千萬張保單，可看到投保人與被保險人的資訊（例如投保年齡、職業、體位、居住地區等）以及保單的資訊（例如保額、保費、繳別、佣金率、通路、附約），資料期間是1994~2014。 我們將於計畫期間的第一年運用這個資料庫探討：有哪些個人特質與商品特徵的變數會影響死亡的機率。過去的文獻多是用整體人口的樣本資料，集中於探討個人特質的影響；此計畫則是運用投保人/被保險人的資料，其所含有的個人資料與商品特質比文獻中所使用的詳盡得多。預計遵循文獻採用離散型的Cox regression估計存活機率，估計的方法將為最大概似法。 第二年的計畫則是運用此資料庫建構死亡率的推估模型。由於此資料庫的期間還不夠長，因此我們的第一步是先運用全國死亡率的資料，建構因子模型，再對所找出來的因子建構時間序列模型，即可進行全國死亡率的推估。在建構時間序列模型時，我們將根據文獻加入一些總體經濟的變數。第二步則是從資料庫中整理出各歷年度的死亡率表/曲線。第三步則是建立公司的死亡率曲線和全國曲線之間的關係。我們將採用關連性模型（Relational Modeling）以及Panel Regression來建立這兩者之間的關係。 第三年的計畫則是運用多群因素分析法於所整理出來的七類商品的死亡率曲線以及全國的死亡率曲線。預計能找出跨群的共同因子以及分屬個別群的獨特因子。透過這樣的分析，我們可以更進一步知道第二年計畫中所找出的總經變數對個別商品影響力之差異；而所找到的獨特因子將對商品間的風險衡量與管理有顯著的意義。 由於本計畫所建立的模型是使用公司的內部資料，建好的模型也將給公司使用，因此算是公司的內部模型。此公司未來也將使用這些模型來估計死亡率的風險，進而估算出死亡率的經濟資本。
dc.description.abstract (摘要) We plan to build three mortality models based on the proprietary information provided by a major Taiwanese company. Currently, the database comprised around ten million contracts, including all kinds of whole life insurance contracts. This database includes detailed information of policyholder and insured such as age, occupation, residential area, and BMI, and also the information of each contract such as coverage, premium, commission, distribution, and rider with the main contract. The sampling period starts from 1994 to 2014 and still expanding. In the first year of this project, we investigate the relationship between mortality rate and insured’s personal and contractual characteristics. Past literature mostly focused on the personal aspect, with aggregate data. Our project capitalizes on the much more detailed insured data than the past literature. Following the previous literature, we use the Cox regression to find the variable that explains the differential mortality rate. We estimate the model with maximum likelihood. The variable we identified will affect the risk of premium, reserve, and contract in general. The goal of second-year project is to build a model that allows us to forecast the mortality rate. The length of dataset is not long enough to support a forecasting model on its own. Therefore, the first step is to estimate a mortality model using the national population data. Then we estimate a relational model that connects the insured population to the national population so that we can extrapolate the dynamic to insured population. The third year’s goal is to analyze insured population and general population with multi-group factor analysis. We expect to find some common factors that exist among all groups and unique factors that belong to each individual group. Our analysis will focus on the impact of variable we found in the second year project to each insurance product. The unique factor we found would provide interesting information to risk management and pricing of each product. Because the data we used is provided by the life insurance company, we will provide our model to the company as an internal model. This is why they are willing to provide us the data. They will use these models to estimate the mortality risk and economic capital.
dc.format.extent 116 bytes-
dc.format.mimetype text/html-
dc.relation (關聯) 科技部, MOST105-2410-H004-070-MY3, 105.08-108.07
dc.title (題名) 建立內部模型管理壽險公司的死亡率風險
dc.title (題名) Managing a Life Insurer’S Mortality Risk by Building Internal Models
dc.type (資料類型) report