本文比較三種估計風險值的模型(常態分配、T分配和Cornish-Fisher展開式)，並考慮變異數-共變異數中三種不同的波動估計(SMA、EWMA和GARCH(1,1))，應用分析在4種信賴水準(99.5%, 99%, 95%,90%)與3種評估原則(保守性、正確性以及效率性)下的績效，提供使用者選擇較佳的風險值預測模型。本研究以倫敦波羅的海交易所（Baltic Exchange in London）中的波羅的海乾散貨指數(BDI)、波羅的海海岬型(Baltic Capesize Index，BCI)和波羅的海巴拿馬極限型(Baltic Panamax Index，BPI)等3種指數為研究對象。 實證研究顯示若以巴賽爾銀行監理委員會規定內部模型之正確性評估準則，Student T分配下之分位數法相較於常態分配與Cornish-Fisher展開式的估計為佳。此外，本文之實證亦顯示風險值的估計分配模型和預測波動模型之選擇會影響風險值預測之正確性。 This article aims to provide users with applicable Value-at-Risk (VaR) models by accommodating three types of quantiles and volatility estimations, examining VaR model performance at four confidence levels with three criteria-accuracy, efficiency and conservativeness. The sample is composed of Baltic Dry Index (BDI), Baltic Capesize Index (BCI) and Baltic Panamax Index (BPI) constructed by London's Baltic Exchange. Based on Kupiec's (1995) unconditional coverage test in the internal model analysis regularized by the Basel Committee, the VaR model estimated with Student's T distribution is verified to be the most accurate measurement compared with variance-covariance method and Cornish-Fisher expansions. Furthermore, the empirical evidence shows that the quantile types and volatility proxies influence the accuracy of VaR models significantly.
