楊朝成Yang, Chau-Chen臺灣大學:財務金融學研究所胡羚Hu, LingLingHu2010-05-112018-07-092010-05-112018-07-092009U0001-2505200920593100http://ntur.lib.ntu.edu.tw//handle/246246/182851天然災害發生的頻率低，但是一旦災害發生就會導致巨額的損失與付出龐大的社會成本。台灣地處全球颱風影響地區之一，其多山的地形與豐沛的降雨量更使得台灣成為颱洪泛濫的危險地區。因此，颱洪的損失管理便成為政府與保險公司最重要的問題，使用正確的損失模型與風險測量指標以評價出精確的價格與承擔償債能力是刻不容緩。文分兩部份：第一、空間綜橫模型用於分析台灣23縣市每年的損失。實證顯示空間效果確實存在，其藉由增加空間效果在綜橫模型中，會比以往的計量模型更具有解釋能力。第二、在此颱洪損失模型是用貝氏蒙地卡羅馬可夫鏈的方法，過去大多數的研究都利用對數常態分配建立巨災損失模型，事實上許多學者都指出這些模型都有右偏與厚尾現象，本論文就此提出柏拉圖分配是適切的颱洪損失模型並且計算不同的風險測量指標並透過模擬與數值分析證明“標準尾部誤差”指標與 “變形雙指數”指標是比以往指標更適切。本論文提供政府與保險公司適切的評價管理模型。Natural disasters occur with low frequency but often cause tremendous damage and social loss. Located in the typhoon belt, Taiwan is one of the most hazard prone areas in the world. Its mountainous topography and high rainfall increase its risk of flooding. Hence, managing the risk of typhoons and floods is an important issue for Taiwan’s government and insurance companies. Using the wrong loss models or risk measure indexes to price insurance products could lead to inaccurate pricing and insolvency. his dissertation has two parts. First, a spatial panel model was used to analyze the annual loss of twenty-three cities in Taiwan. The empirical results showed that spatial effect existed. By adding spatial effect into panel model, the model can explain more than the fixed effect panel model and random effect panel model. Second, a loss distribution of typhoons and floods was built by using a Bayesian Monte Carlo Markov Chain method. Although most researchers use Log-Normal distributions to model a catastrophic loss, this is not appropriate because the loss distribution has positive skewness and is heavy-tailed. Recently, McNeil and Frey (2000), Rootzen and Tajvidi (2000), Thuring, Gustafsson and Pritchard (2008), also stated that using Log-Normal distribution to model a characteristic loss is not appropriate. Our results showed that the Pareto distribution is more adaptive for modeling a typhoon and flood’s loss. Finally, we evaluated different risk measure indexes through simulating and numerical analysis. The simulation results demonstrated that “Tail Standard Deviation” and “Dual Power Distortion” are more suitable while considering the tail loss risk. This dissertation aims to provide guidance to insurance companies and governments in assessing and managing typhoon and flood risks.口試委員會審定書 i謝 ii文摘要 iv文摘要 v. Introduction 1.1 Overview of Natural Disaster Risk around the Globe 1.2 The Typhoon and Flood Losses in Taiwan 6.3 Loss Distribution Modeling and Risk Measure Indexes 7. Panel Models of Typhoon and Flood Losses 10.1 Traditional Panel Models 11.2 Test for Model Specification 16.3 Panel Models with Spatial Correlation 19.4 Measuring Spatial Association and Correlation 24. Spatial Panel Model Empirical Results 28.1 Data Description and Hypotheses 29.2 Empirical Results to Traditional Panel Models 33.3 Empirical Results to Spatial Models 36. Typhoon and Flood Losses Distribution 39.1 Typhoon and Flood Losses Distribution Modeling 39.2 Catastrophe Risk Measure Index 41. Empirical Data Analysis and Numerical Results 47. Conclusions 54eferences 55ppendix 60application/pdf719000 bytesapplication/pdfen-US空間綜橫模型颱洪損失分配蒙地卡羅馬可夫鏈法風險測量指標Spatial Panel ModelTyphoon and flood lossesMonte Carlo Markov ChainRisk Measure Index[SDGs]SDG11thesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/182851/1/ntu-98-D90723007-1.pdf