臺灣商業火險平均費率對全球巨災再保險費率指數變動之反應

Abstract

本研究以臺灣商業火災保險及其附加險之市場年度資料，結合全球巨災再保險費率指數，實證檢驗再保成本變動對商業火險平均費率之當期傳導程度，並評估價格調整是否具有不對稱性。再保成本以 Guy Carpenter Global Property Catastrophe Rate-on-Line（ROL）Index 衡量；商業火險平均費率則以「滿期保費／保險金額」建構，並以對數差分衡量年際變動。樣本期間為1996–2022年，主要估計期間為1997–2022年（26筆年度資料觀察值）。實證模型採用再保成本變動之正負拆分規格，並以重大損失指標（損失率達樣本第90百分位數）控制事件年衝擊；同時以非規章費率保費佔比作為費率自由化實際開放程度之代理，檢驗不同市場結構下之異質性。在推論方法上，本文採用 Newey–West（1987）HAC 標準誤，針對 OLS 迴歸係數估計量之變異數進行修正，以降低殘差存在異質變異與一定程度自我相關時，對標準誤與顯著性判斷所造成之影響。另外，考量本文樣本期間有限，傳統漸近推論於小樣本下可能產生偏誤，本文進一步採用結合區塊重抽樣之wild bootstrap 建構有限樣本下之 p 值與信賴區間。此方法結合 wild bootstrap 對異質變異之穩健性，以及 block bootstrap 對時間序列相依結構之保留，以提升小樣本統計推論之可靠性。 結果顯示，在年度頻率下，再保成本上升與下降之當期傳導係數雖多為正，但均未達統計顯著，對稱性檢定與「上漲較快、下跌較慢」之方向性檢定亦未獲支持，顯示不易在小樣本年度資料中穩健辨識再保成本之即時線性傳導。相較之下，重大損失指標與平均費率變動呈現較一致之正向關聯，且其影響主要集中於非規章佔比較高的年度；全樣本交互項模型亦證實重大損失在高非規章環境下之額外調價幅度顯著較大。 整體而言，臺灣商業火險平均費率之調整更呈現「事件年主導」，且具有一定程度之平滑化與延遲性，而非逐年緊貼再保成本訊號。本文據此提出監理與實務意涵：在維持資本適足與資訊揭露之前提下，可針對大型風險建立更具風險敏感度之定價彈性與事後回溯機制，以強化費率信號與風險管理投資之連結。

This thesis examines whether, and to what extent, changes in global catastrophe reinsurance costs are passed through to the pricing of commercial fire insurance in Taiwan at an annual frequency. Using market-level data on commercial fire insurance and its endorsements in Taiwan, this study links domestic premium dynamics to the Guy Carpenter Global Property Catastrophe Rate-on-Line（ROL）Index, which serves as a proxy for reinsurance costs. The average premium rate is constructed as earned premiums divided by insured amounts, and annual changes are measured using log differences. The sample period spans 1996–2022, with the estimation period covering 1997–2022, yielding 26 annual observations. The empirical specification decomposes changes in reinsurance costs into positive and negative components in order to test for asymmetric adjustment. A big loss indicator, defined as years in which the loss ratio exceeds the 90th percentile of the sample distribution, is included to control for event-year shocks. To capture heterogeneity in the effective degree of rate liberalization, this study uses the share of non-regulated premium as a proxy for market structure and compares high- and low-share regimes. Statistical inference is based on OLS estimation with Newey–West（1987）heteroskedasticity- and autocorrelation-consistent（HAC）covariance estimation. Given the limited sample size, finite-sample block wild bootstrap p-values and confidence intervals are further employed. The empirical results provide limited evidence of contemporaneous pass-through from changes in reinsurance costs to the average premium rate at the annual horizon. Neither the symmetry test nor the directional “rockets-and-feathers” hypothesis is supported. By contrast, the big loss indicator is consistently associated with increases in the average premium rate, and this effect is significantly stronger in years with a higher share of non-regulated business. The pooled interaction model further confirms the presence of cross-regime differences. Overall, the pricing of commercial fire insurance in Taiwan appears to be driven more by discrete event-year adjustments, accompanied by smoothing and stickiness, than by a stable and immediate linear pass-through of global reinsurance cost shocks. The policy implications of this study highlight the importance of enhancing risk-sensitive pricing flexibility for large corporate risks, while ensuring transparent pricing governance and establishing ex post review mechanisms.