次貸風暴所引發的連鎖效應對全球金融市場造成巨大的衝擊，也讓全球重新省思風險管理的價值。其中又以美國國際集團(AIG)的破產事件影響最鉅，因此本文以台灣某產險公司的汽車車體損失險為研究對象，評估該險種損失分配與相依結構。首先利用損失因子對損失資料做交叉分析與分量迴歸分析；再利用一般化柏拉圖極端值模型(GPD)分別計算95%、97.5%與99%信賴水準下的風險值，透過拔靴複製法來估計其信賴區間並檢測模型績效；最後使用四種copula模型來檢測甲、乙與丙式汽車車體損失險之間的關聯性。實證結果顯示，GPD模型能夠精準的配適損失資料的尾端部分，在99%的信賴水準為下，預期損失為469409.6738，損失超過50萬的機率為0.25%。甲、乙與丙式險種之間的相依結構極低，隱含三式車險之間彼此獨立。 After the 2007 finance crisis, the risk management is more noticeable in recent years, but there is still very limited number of general literatures on automobile physical insurance. This paper focuses on modeling and estimating tail parameters of automobile physical damage loss severity. At an attempt to do so, firstly, the cross analysis and quantile regression are used to examine the correlations between risk factors and loss. Then, we proceed with a simple exploratory loss data analysis using Q-Q plot and cross analysis. Furthermore, we determine the thresholds of GPD through mean excess plot and Hill plot. Value at Risk and the expected shortfall are also calculated. Bootstrap method is taken into account to estimate the confidence interval of parameters. Empirical results show that the GPD method is a theoretically well supported technique for fitting a parametric distribution to the tail of an unknown underlying distribution. Copula functions are also applied to fit the rank correlation between different loss types. From the results, it is concluded that the GPD model can capture the behavior of the loss severity tail of automobile physical damage insurance very well.