在風險管理中，風險值為廣泛使用來衡量市場風險的指標。適當的風險值應有效地測度超限事件的發生，但業界所廣泛使用的歷史模擬法卻無法有效捕捉價格的跳點，使得在跳點發生較頻繁期間，超限事件具有叢聚性之特性。本文提出以自我迴歸條件時距模型 (ACD 模型) 來預測跳點的發生，並在跳點可能發生的期間調整風險值，來改善歷史模擬法所估計之風險值。基於波動性模型擅於刻劃報酬叢聚性之特性，本研究同時考慮以 Realized 波動度、 Bi-power 波動度、 GARCH 模型來估計風險值，並以超限率和資本計提之嚴謹程度的觀點與調整後的歷史模擬法比較模型之適應性。結果發現，在同時考慮超限事件以及跳點警訊下，針對歷史模擬法進行調整的確能有效降低超限事件的發生至目標水準，且其資本計提之嚴謹性也不比波動性模型來的差。 In risk management, value-at-risk is widely used to measure the market tail risk. Value-at risk should measure the violations effectively, but the VaR calculated by historical simulation may experience a sequence of cluster of violations for its failing to accommodate the price jump behavior. This paper applies the autoregressive conditional duration model ( ACD model ) to predict jump durations, and adjust the VaR based on historical simulation using the predicted jump arrival time to reduce the violation rate. Since the volatility models have the strength of capturing the clustering of returns, we also consider the VaR calculated by the Realized volatility, Bi-power volatility and GARCH models. In view of violation rate and prudentiality of capital requirement, we then compare the VaR from the adjusted historical simulation with those from these volatility models. The results show that by properly modeling the jump durations and arrival of violations to adjust historical simulation can significantly reduce the violations to target level. Moreover, it leads to a better prudentiality of capital requirement no worse than that from the volatility models.
