根據以往探討評價合成型抵押擔保債務憑證之文獻研究，最廣泛使用的方法應為大樣本一致性資產組合(large homogeneous portfolio portfolio；LHP)假設之單因子常態關聯結構模型來評價，但會因為常態分配的厚尾度及偏斜性造成與市場報價間的差異過大，且會造成相關性微笑曲線現象。故像是Kalemanova et al.在2007年提出之應用LHP假設的單因子Normal Inverse Gaussian(NIG)關聯結構模型以及邱嬿燁(2007)提出NIG及Closed Skew Normal(CSN)複合分配之單因子關聯結構模型(MIX模型)皆是為了改善其在各分劵評價時能達到更佳的評價結果 ，然而過去的文獻在評價合成型抵押擔保債務憑證時，需要將CDS價差、各分劵真實報價之資訊導入模型，並藉由此兩種資訊進而得到相關係數及報價，故靜態模型大多為事後之驗證，在靜態模型方面，我們嘗試使用不同概念之CDS取法以及相對到期日期數遞減之概念來比較此兩種不同方法與原始的關聯結構模型進行比較分析，在動態模型方面，我們應用與時間序列相關之方法套入以往的評價模型，針對不同商品結構的合成型抵押擔保債券評價，並由實證分析來比較此兩種模型，而在最後，我們利用時間序列模型來對各分劵進行預測。

根據以往探討評價合成型抵押擔保債券之文獻研究，最廣為使用的方法應用大樣本一致性資產組合(large homogeneous portfolio portfolio ; LHP)假設之單因子常態關聯結構模型來評價，但會造成合成型抵押擔保債券憑證與市場報價間的差異過大，且會造成相關性微笑曲線現象。由文獻顯示，單因子關聯結構模型若能加入厚尾度或偏斜性能夠改善以上問題，且對於分券評價時也會有較好的效果，像是Kalemanova et al. (2007) 提出應用LHP假設之單因子Normal Inverse Gaussian(NIG)關聯結構模型以及邱嬿燁(2007)提出NIG及Closed Skew Normal(CSN)複合分配之單因子關聯結構模型(MIX模型)在實證分析中得到極佳的評價結果。自2008年起，合成型抵押擔保債券商品結構開始出現變化，而以往評價合成型抵押擔保債券價格時，商品結構皆為同一種型式。本文將利用常態分配、NIG分配、CSN分配以及NIG與CSN複合分配作為不同的單因子關聯結構模型，藉由絕對誤差極小化方法，針對不同商品結構的合成型抵押擔保債券評價，並進行模型比較分析。由最後實證分析結果顯示，單因子NIG(2)關聯結構模型優於其他模型，也證明NIG分配的第二個參數 β 能夠帶來改善的評價效果，此項證明與過去文獻結論有所不同，但 MIX模型則為唯一一個符合LHP假設的模型。 / Based on the literature of discussing the approach for pricing synthetic CDOs, the most widely used methods used application of Large Homogeneous Portfolio (LHP) assumption of the one factor Gaussian copula model, however , it fails to fit the prices of synthetic CDOs tranches and leads to the implied correlation smile. The literature shows that one factor copula model adding the heavy-tail or skew can improve the above problem, and also has a good effect for pricing tranches such as Kalemanova et al (2007) proposed the application of LHP assumption of one factor NIG copula model and Qiu Yan Ye (2007) proposed the application of LHP assumption of one factor NIG and CSN copula model. This article found that the structure of synthetic CDOs began to change since 2008. The past of pricing synthetic CDOs, the structure of synthetic CDOs are the same type, so this article will use different one factor copula model for pricing different structure of synthetic CDOs by using the absolute error minimization. This article will observe whether the above model can be applied in the new synthetic CDOs and implement of different type model for comparative analysis. The last empirical analysis shows that one factor NIG (2) copula model is superior to other models, more meeting the actual market demand, also proving the second parameter β of the NIG distribution able to bring about improvements in pricing results. This proving is different for the past literature conclusions. However, the MIX model is the only one in line with the LHP assumptions.

We model Normal Inverse Gaussian distributed log-returns with the assumption of stochastic volatility. We consider different methods of parametrization of returns and following the paper of Lindberg, [21] we assume that the volatility is a linear function of the number of trades. In addition to the Lindberg’s paper, we suggest daily stock volumes and amounts as alternative measures of the volatility. As an application of the models, we perform Value-at-Risk and Expected Tail Loss predictions by the Lindberg’s volatility model and by our own suggested model. These applications are new and not described in the literature. For better understanding of our caluclations, programmes and simulations, basic informations and properties about the Normal Inverse Gaussian and Inverse Gaussian distributions are provided. Practical applications of the models are implemented on the Nasdaq-OMX, where we have calculated Value-at-Risk and Expected Tail Loss for the Ericsson B stock data during the period 1999 to 2004.

Zur Approximation der Dichte von linearen Integralfunktionalen schwach korrelierter Prozesse mit Korrelationslänge wurden bisher Gram-Charlier-Reihen benutzt. In diesem Artikel werden weitere Verfahren zur Dichteapproximation "integraler" Zufallsgrößen beschrieben und untersucht, ob sie sinnvoll auf das Dichteapproximationsproblem bei linearen Integralfunktionalen angewendet werden können.

This essay investigates how realized variance affects the GARCH-models (GARCH, EGARCH, GJRGARCH) when added as an external regressor. The GARCH models are estimated with three different distributions; Normal-, Student’s t- and Normal inverse gaussian distribution. The results are ambiguous - the models with realized variance improves the model fit, but when applied to forecasting, the models with realized variance are performing similar Value at Risk predictions compared to the models without realized variance.

在評價合成型抵押擔保債券憑證時，需考慮多個標的資產間之違約相關性。根據過去評價合成型抵押擔保債券的文獻研究，發展高斯分配等單因子關聯結構模型，在給定LHP假設之下，執行各分券評價時，僅有在權益分券(equity tranche)得到好的配適結果，還會造成相關性微笑曲線(correlation smile)等問題。文獻研究，單因子關聯結構模型若能加入厚尾度或偏斜性能夠改善以上問題，且對於分券評價時也會有較好的效果，像是Kalemanova et al. (2007)提出應用LHP假設之單因子NIG關聯結構模型，或是Dezhong et al. (2006)提供之單因子關聯結構延伸模型，來評價抵押擔保債權憑證。進一步發現，全世界主要的信用違約指數的標的資產個數不一，最少有14個標的資產(CDX.EM)，最多有125個標的資產(iTraxx Europe)，事實上標的資產個數均不多，而過去文獻常建立在大樣本假設下進行抵押擔保債券之評價，本文研究目的在於，針對單因子高斯關聯結構模型，建立單因子高斯關聯結構延伸模型，假設在非大樣本性質下，評價合成型抵押擔保債券憑證，嘗試觀察是否有較佳的估計結果，改善相關性微笑曲線的現象。本文將利用常態分配、NIG分配以及非大樣本之常態分配作為不同的單因子關聯結 構模型，藉由絕對誤差極小化方法，針對不同商品結構的合成型抵押擔保債券評 價，並進行模型比較分析。實證結果顯示，非大樣本之常態分配關聯結構模型與LHP假設下的單因子高斯關聯結構模型有類似的評價結果，但在近兩年(2012年、2013年)的實證分析結果顯示，非大樣本之常態分配關聯結構模型於前四分券評價結果上符合同質性假設，即各個資產對共同因子的相關性近乎相同。

因為股票市場常具有厚尾、偏態和峰態的特性且在國際的股票市場之間，股票報酬長存在有尾端相依的情況，所以我們的資產模型不能選用Gaussian分配。 近幾年來，常用GH 分配建構單維度的股票報酬。這篇文章將利用多元仿射JD、多元仿射VG 和多元仿射NIG分配去建構風險性資產的報酬並請應用到資產配置。 建構風險性資產的報酬後，我們提供兩種不同形式的投資組合並且可以導出投資組合的期望值、變異數、偏態和峰態。我們嘗試以投資組合的期望值、變異數、偏態和峰態當成我們的目標函數，然後得出未來最佳的投資組合的權重。為了讓我們的資產配置更加動態和有效率，我們重新估計模型的參數、選擇最佳的投資組合權重，然後重新評估最佳的資產配置在每個決策日期。實證結果發現當股票市場的表現好的時候，我們建議資產配置應使用偏態當成我們的目標函數，但是當股票市場的表現太好的時候，我們建議資產配置應使用變異數當成我們的目標函數。 / Since the stock markets always have the characteristics of heavy-tailness, skewness and kurtosis and there exists tail dependence among the international stock markets, we can’t use the Gaussian distribution as our model. Recently, the generalized hyperbolic (GH) distribution has been suggested to fit the single stock returns. This article will use the multivariate affine JD (MAJD), multivariate affine variance gamma (MAVG) and multivariate affine normal inverse Gaussian (MANIG) distributions to construct the risky asset returns, and apply them to asset allocation. After constructing the risky asset returns, we provide two different forms of portfolio and obtain the mean, variance, skewness, kurtosis of portfolio. We can try to select the optimal weights of portfolio by using the mean, variance, skewness, kurtosis of portfolios as our objective functions. To make our asset allocation more dynamic and efficient, we re-estimate all parameters for our models, select the optimal weights of portfolio, and re-assess the optimal asset allocation at each decision date. Empirically, when the performances of stock markets are good, we suggest that our asset allocation uses the skewness as the objective function. When the performances of stock markets are not good, we suggest that our asset allocation uses the variance as the objective function.