AV-GARCHM模型於金融控股公司市場風險值之研究

Abstract

本文採用AV-GARCHM模型伴隨不同的報酬結構－ARMA(1,1)、AR(1)、MA(1)、In-Mean－來檢測其於金融控股公司VaR預測值上的表現。我們沿用2003年王所模擬的兩個投資組合，以逐日計價原則計算每日P&L值. 並據此分別估算99%與95%信賴水準下的領先一天VaR預測值。在違約次數的合格性與資本提列有效性的考量下，本研究有以下發現： 1. 在99%信賴水準下，四個VaR預測模型都只發生一次違約，小於 Basel規定的2次；在95%信賴水準下，四個VaR預測模型在投資組合A都產生2個違約次數而投資組合B僅產生1個違約次數；以損失的超出次數為基準，此四個VaR預測模型均可視為合格的內部VaR市場風險模型。 2. AV-GARCHM模型同時考慮了訊息的平移與旋轉效果，理論上，我們假設其應當優於僅考慮旋轉效果的EGARCHM模型以及僅考慮平移效果的NA-GARCHM模型。然而在既定的模擬組合與資料期間下，除了ARMA(1,1)模型，我們並無法明確判斷其他三個模型相對於EGARCHM以及NA-GARCHM模型優越。

In this paper, we employ the AV-GARCHM model with various mean equations to evaluate their performance as VaR forecast models. We form two simulated portfolios, and calculate their daily profit and loss based on marking to market rule. Forward testing of one-day-head VaR models under 99% and 95% confidence level is evaluated with realized P&L of two simulated portfolios. Based on the consideration of violation number and capital charge efficiency, we have the following findings: 1. All of the four models generate only 1 violation number under 99% confidence level and 2 violations in portfolio A and 1 violations in portfolio B under 95% confidence level. 2. AV-GARCHM model considers both shift and rotation effect to news shock. Theoretically, we assume it should be better than EGARCHM model and NA-GARCHM model. However, except ARMA (1, 1) model, all rest models perform equally better in terms of violation number in both portfolio A and portfolio B. Thus, we cannot say that AV-GARCHM model is absolutely better than EGARCHM model or NA-GARCHM model.