Threshold-GARCH 模型於金融控股公司市場風險值之研究

Abstract

由於金融自由化以及全球化的趨勢，金融機構從事金融交易活動的廣度與深度都呈現增加的態勢。因此，金融機構本身持有的投資組合不僅金額愈來愈大，其組成結構也愈來愈複雜。為增加金融機構本身的競爭力，風險管理的議題也愈來愈受到重視。 本研究藉由Threshold-GARCH模型，來估算國內兩家金融控股公司的市場風險值，並探討將投資組合的變異數或標準差納入報酬率方程式，是否可以更精確地衡量金融機構的市場風險值。由於投資組合的每日交易明細及損益資料屬於較機密的內部資料，因此我們透過模擬的方式，模擬出投資組合A和B，分別代表國內兩家金融控股公司的投資組合。模擬的持有期間從2000年11月28日至2003年4月15日，共617個實際交易日。我們利用前400筆資料進行參數的估計，之後對後217個交易日的市場風險值進行預測，並且依據預測結果做模型比較，尋找預測能力較佳的模型。 我們發現Threshold-GARCH模型對於兩個金融控股公司的市場風險值有良好的預測能力，同時將變異數或標準差納入報酬方程式時，也提供較有效的預測。此外，似乎市場對好壞消息的反應是不對稱的。然而，我們也發現有些模型預測較低的資本計提但卻須承擔較多每日損益超過市場風險值的次數，因此必須在兩者之間有所取捨。

Due to deregulation and globalization, financial institutions can do diversiform activities. Therefore, trading accounts in financial institutions not only have become larger but also have become much more complex than before. Risk management becomes an important issue because of competitiveness in financial markets. We adopt Threshold-GARCH model to forecast VaR of two financial holding companies in Taiwan. And we check if the forecasting results are more accurate when variance or standard deviation is included in the mean equation. However, trading data of financial holding companies are highly confidential and are not publicly accessible. To solve the problem mentioned above, we conduct our models based on two simulated portfolios. The holding period of simulated portfolios is from 2000/11/28 to 2003/4/15. We use first 400 observations to estimate parameters and compare the forecasting results with the rest of observations. We find that Threshold-GARCH model performs well in VaR forecasting whether variance or standard deviation is included in the mean equation or not. And news impact seems to be asymmetric in our testing results. We also find that there is a trade-off between market risk capital charge and the number of violations.