風險值與風險管理策略之研究

VaR and Risk Management Strategy

林淑蓉(Shu-Long Lin)

指導教授 ： 陳錦村

國立中央大學/管理學院/財務金融學系/碩士(2006年)

摘要

本研究主要探討風險值模型與風險管理政策兩個層面：（一）從保守性、準確性與效率性三個面項，比較歷史模擬法、共變異數法、蒙地卡羅法－多變量常態分配、蒙地卡羅法－固定條件相關係數模型(Constant Conditional Correlation, 簡稱CCC)與蒙地卡羅法－動態條件相關係數模型(Dynamic Conditional Correlation, 簡稱DCC)五種風險值的預測能力；（二）在不同停損機制下，動態模擬2005年股票、外匯與債券個別投資組合，並以風險調整後報酬(RAROC)比較不同停損機制的投資組合績效。本文資料來源為台灣經濟新報（簡稱TEJ）、台灣櫃臺買賣中心(OTC)與全球財經資料庫(Datastream)，研究期間為2001年1月1日至2005年12月31日。
五種風險值模型在保守性、準確性與效率性的表現隨資產種類不同而有
所不同。就股票而言，綜合回顧測試中的各項指標：DCC模型估計的風險值，準確性與效率性優於其他四種模型，特別是在覆蓋損失能力上，DCC估計的風險值覆蓋損失的能力接近100%，所以DCC模型配適能力最高。外匯部分，準確性則以歷史模擬法表現最為突出；效率性則是以共變異數法最佳。債券方面，仍以歷史模擬法保守性；準確度以蒙地卡羅法－多變量常態表現最佳，其他風險值方法表現差強人意，可能原因是參數估計期不具代表性，2001年1月至2004年12月五年公債利率由5%多掉到1%多，以此段期間當作估計期，將高估債券報酬波動率，以致於歷史模擬法、CCC與DCC模型過度高估風險值。實證結果顯示：資產報酬不同，其報酬分配與特性也不同，適用的風險值模型也不同。
? 另外，動態模擬的投資組合，僅股票觸及門檻率，進行資產調整的動作。每週觀察，20%門檻率的投資組合RAROC最高。敏感度分析中，不同停損機制的投資組合對於報酬變動的敏感度差異不大。就風險模型而言，在股票、外匯與債券的投資組合中，皆以CCC與DCC模型估計的風險值，對於報酬波動率的敏感度最高。

關鍵字

CCC ； 風險管理 ； 風險值 ； DCC

並列摘要

The paper focuses on two things：First, to compare conservation, accuracy, and efficiency of forecast performance of 5 VaR models：historical simulation method, Variance-Covariance method, Monte-Carlo simulation method(Multi-Normal), Monte-Carlo simulation method(Constant Conditional Correlation, CCC), and Monte-Carlo simulation method(Dynamic Conditional Correlation, DCC)；Second, use RAROC to compare stock portfolio performances simulated by different stop-loss methods, and so as foreign exchange portfolios and bond portfolios. Data are derived from Taiwan Economic Journal database, Gre-Tai Market (OTC), and Datastream database；sample period begins from 2001 to 2005.
Conservation, accuracy, and efficiency of five VaR methods perform differently form stock, foreign exchange, to bond portfolios. For stock portfolios, DCC method performs best in accuracy and efficiency. As for foreign exchange portfolios, historical simulation method stands out both in conservation and accuracy； however, variance-covariance method performs best in efficiency. As far as bond portfolios are concerned, Multi-Normal method performs best in conservation, accuracy, and efficiency, whereas forecasts of the other four methods tend to so conservative that number of these model’s exception are zero which may result form selection of sample period. In 2001, yield to maturity of 5 year government bond is around 5% which yield to maturity drops to 2% or so in 2004；nevertheless, yield to maturity of 5 year government bond fluctuates around 1.7%；Therefore, an extremely variable sample period would lead to overestimate variation of bond returns which result in model misspecification. Empirical evidence shows that the optimal VaR method differs from asset returns.
Concerning dynamic simulation portfolios, only stock portfolios have ever touched
hurdle rate and adjusted portfolio holdings. As for RAROC, portfolios observed weekly and hurdle rate is 20% performs best. For sensitivity analysis, CCC method and DCC method’s scores are highest and that means CCC method and DCC method whose capacity of covering loss better than the other three methods.

並列關鍵字

DCC ； Risk management strategy ； CCC ； VaR

參考文獻

5、高櫻芬、謝家和 (2002)，「涉險值之衡量－多變量GARCH模型之應用」，經濟論文

6、 劉美纓、吳俊賢、吳壽山 (2003)，「銀行市場風險與適足資本之研究—內部模型法」，

1、Bollerslev, T. (1990), “Modeling the Coherence in Short-Run Nominal Exchange Rates:

Multivariate Generalized ARCH Approach,” Review of Economic and Statistics.

derivatives and risk management.