氣候風險對台灣產險公司的影響： ACI – Copula 壓力測試之實證研究

Abstract

近年極端氣候事件發生的頻率與災害強度顯著上升，氣候風險已成為金融體系潛在之系統性風險來源。相較於多數文獻聚焦於資產端曝險與情境分析，本研究從負債端出發，探討氣候變遷對臺灣產險公司理賠損失分布尾端風險與相依結構的影響。 本研究以2002年1月至2024年12月臺灣市占前八家產險公司之月資料為樣本，涵蓋9類主要險種的已發生賠款金額，並建構臺灣地區的精算氣候指標(Actuaries Climate Index, ACI）。方法上，首先以ARMA-GARCH模型處理邊際分布之厚尾與波動叢聚特性，接著採用C-vine Copula刻劃ACI與各險種損失間之非線性與尾端依賴結構，並透過條件模擬進行情境壓力測試，估計不同氣候強度下之條件損失分布與99.5% VaR變化。 實證結果顯示，臺灣產險損失對氣候風險的反應具有顯著的狀態相依特徵。在平均層面氣候變數與損失的相關性有限，但於極端情境下尾端依賴顯著上升，並呈現非線性與不對稱結構。條件模擬進一步指出，極端氣候情境將導致損失分布右移與門檻式放大，99.5% VaR 普遍上升，且資本壓力之差異主要來自險種結構與曝險集中度，而非公司規模，此結果顯示氣候風險主要透過極端尾端機制影響產險體系，而非透過整體平均相關結構反映。 綜合以上，本研究顯示氣候風險主要透過極端尾端機制影響產險體系，支持監理架構由平均風險評估轉向高分位情境導向的資本壓力測試設計，以強化保險體系對極端氣候衝擊之韌性。

In recent year, the frequency and intensity of extreme climate events have increased significantly, and climate risk have become potential source of systemic risk to the financial system. While the majority of existing studies focus on asset-side exposures and scenario-based analyses, this study focus on liability side, discussing how climate change affects the tail risk and dependence structure of incurred losses in Taiwan’s property and casualty (P&C) insurance industry. This study uses monthly data on incurred losses across nine major insurance lines from January 2002 to December 2024, focusing on the eight largest Taiwanese P&C insurers by market share, and construct a localized Actuaries Climate Index (ACI) for Taiwan. Methodologically, we first employ ARMA-GARCH models to capture the heavy-tailed distributions and volatility clustering in the marginal loss processes. Subsequently, we employ a C-vine copula framework to model the nonlinear and tail dependencies between the ACI and individual insurance line losses. Using this framework, we conduct conditional simulations to perform climate stress testing, estimating scenario-dependent loss distributions and the resulting fluctuations in the 99.5% Value-at-Risk (VaR). Empirical results show that the response of Taiwanese P&C insurance losses to climate risk exhibits significant state-dependent characteristics. While average correlations between climate variables and losses remain limited, tail dependence increases significantly under extreme scenarios, exhibiting nonlinear and asymmetric structures. Conditional simulations further indicate that extreme climate scenarios lead to a rightward shift and threshold-based amplification of loss distributions. Consequently, the 99.5% VaR increases across the board. The disparity in capital pressure primarily stems from the business line mix and exposure concentration, rather than firm size. These results show that climate risk affects the P&C sector mainly through extreme tail mechanisms rather than through changes in mean-level dependence. In summary, this study demonstrates that climate risk affects the P&C insurance system primarily through extreme tail mechanisms. These findings support a regulatory transition from average-risk assessments toward high-quantile, scenario-based capital stress testing, thereby enhancing the insurance system's resilience against extreme climate shocks.