運用資料探勘技術在產險行銷之個案研究

A Research of Data Mining Technology Applied to Marketing for Property Insurance

吳長洲(Chang-Chou Wu)

指導教授 ： 林震岩

中原大學/商學院/企業管理研究所/碩士(2004年)

https://doi.org/10.6840/cycu200400202

摘要

國內產物保險之費率自由化已於民國91年4月1日啟動，隨著加入WTO及自由化的腳步，汽車保險費率中的危險保費（或稱純保費）亦將逐步解除管制，在激烈競爭的內外在環境下，保險商品設計及保險價格的多樣化是爾後之必然趨勢，而附加保費則會因各公司之不同經營條件而有更大的調整及設計空間。國內學者針對汽車保險之費率結構及影響因子已有諸多探討，但限於實證資料取得不易，故多針對單一險種、特定變數或局部地區加以研究，或採用抽樣方式而存在抽樣誤差，以致較無法提供完整之論述。
本文著眼於此，取得與全國母體相似並具代表性之個案公司汽車保險之承保及理賠資料共233,579筆，探討影響肇事次數及肇事金額之相關風險因素，並將實證結果與現行費率之實務做法做比較。結果發現影響汽車保險保費的風險因素涵蓋有從車因素與從人因素，其中從車因素除廠牌、使用性質與現行考量的風險因素相符外，並發現費率中尚未考量之排氣量、車齡等因素，與肇事次數及肇事金額均有顯著相關；同時發現現行費率中從人因素僅考量車主部分，與實證結果並不相符，研究結果顯示應加入從人因素中之駕駛人因素（駕駛人性別及年齡），較能符合現況。此外，本研究結果也顯示續保之保戶是較不會肇事的，且駕駛人是否為車主及肇事地區二個變數，對於肇事次數及肇事金額均有顯著區別。接著將篩選出之任意第三人責任險之保單資料共183,882筆導入資料探勘，藉由工具軟體將客戶分群並希望找出目標行銷之客戶特徵。
期望這些結果能提供保險業者於汽車保險商品價格釐訂或修正時，或在針對特定行銷目標設計保險商品時的參考，使被保險人能享有更公平、客觀及更多樣化的保費及商品，達到雙贏的理想。

關鍵字

客戶區隔 ； 汽車保險 ； 風險因素 ； 資料探勘 ； 產物保險

並列摘要

The liberalization of the property insurance premium started on April1, 2002 in Taiwan. With the admission to WTO and the speedy pace of the liberalization, the limitation for the risk premium of automobile insurance, also termed pure premium, would also be released gradually. In the internal and external competitive environment, offering a wide variety of automobile insurance products and various kinds of insurance premium would become a trend. In addition, the loading expense would be adjusted and designed with respect to the operating condition in the insurance company. The local researchers have conducted a lot of studies on investigating the structure of the automobile insurance premium and effecting risk factors. However, the studies on the single insurance or particular areas were limited due to the difficulty of collecting relevant data.
With this in vein, this study aimed to investigate the risk factors which significantly affected the number of the accidents and the payment amount by collecting the acceptance and adjusting data of the automobile insurance cases (a total of 233,579 cases) which represented the national population. Furthermore, the study compared the research results of the study with the current practices on automobile insurance premium. The results showed that the risk factors which affected automobile insurance premium included the vehicles-oriented factors and the person-oriented ones. Regarding vehicle-oriented factors, the factors of automobile brands and the purposes of using the automobile were included in the current risk factors. However, the factors of displacement and the car age were not considered in the current. Both of the displacement factor and car age factor highly correlated with the number of the car accidents and the payment amount. In the current practice of automobile insurance premium, the factor of car owner was considered in terms of the person-oriented factor. That did not correspond to the results of the study. The results indicated that the factors of the age and gender of the car driver should be included. Moreover, the study found that the renewal had fewer car accidents. The other factors contributing to fewer car accidents were the factor of the driver (whether the driver was the car owner or not) and the area of the car accident. By analyzing the total of 183,882 cases of automobile third party liability insurance using the data mining tool, the study classified the clients and discovered the characteristics of the targeted clients.
It is hoped that the findings could provide the insurer with a reference when setting or revising the premium of automobile insurance products or when designing automobile insurance products for particular marketing targets. Therefore, the insured would benefit from fairer and diverse automobile insurance premium and products for the idea of mutual benefits.

並列關鍵字

data mining ； risk factor ； automobile insurance ； property insurance ； customer segmentation

參考文獻

十八. 台灣區車輛工業同業公會，http://www.ttvma.org.tw

一. Romald W. and Edmond L. Escolas (1982) ，1979 Accident Reports: Do Driver Characteristics Support Rate Discrimination? ，Journal of Risk and Insurance, 49 91-103.

三. G. Dionne and C. Vanasse（1989），Automobile Insurance Ratemaking in the Presence of Asymmetrical Information.

五. Dionne, Georges, Christian Gourieroux, and Charles Vanasse (2001)， Testing for Evidence of Adverse Selection in the Automobile Insurance Market，Journal of Political Economy, 109 444-453.

壹. 中文部分：