自用車車體險核保系統之建構－台灣地區之實證研究

汽車車體損失險保費收入一直佔有汽車保險相當高的比例，根據研究，車體損失險，由於單件的平均保費較高，所以對產險公司而言，幾近任意第三人責任險的十倍。而車險損失險的損失率變化也牽動任意車險整體損失率的高低，汽車車體的獲利與否，關係著整體車險之經營結果，也同時決定產險公司的經營績效。
本文應用國內某家具代表性的大型產險公司的汽車車險核保資料，以Logit模型及基因演化類神經網路模型來進行核保系統之建構，實證結果發現此二模型所建構的核保系統評估績效無論樣本內或外均可達百分之八十以上。本文根據此二模型，給予權重，建構核保評分表，供核保實務應用。
關鍵字：汽車車體險，核保系統，Logit模型，基因演化類神經網路模型

The Construction of the Physical Damage Automobile Insurance Underwriting System—Empirical Study in Taiwan

　　　　　　　　　　　　　　　　　　　　　　　　(Wo-Chian Lee)＊(Jason, Ting)

The revenue from physical damage insurance is typically a high proportion of automobile insurance revenue. According to research, physical damage insurance premiums are ten times third-party liability insurance. However, changes in the loss ratio for physical damage insurance can affect other insurance policies. For example, whether physical damage insurance earns profit or not affects the business performance of property insurance companies.
This study examines underwriting data on physical damage insurance from a Taiwanese property insurance company. We use logit and GANN models to construct a hybrid underwriting system. Empirical results show that the two models estimate over 80% of true rates whether training sample or test sample data is used. Finally, we construct an underwriting system by weighting the two models.
Keywords: automobile material damage insurance, logit model, GANN model