近幾年來實證研究發現金融資產報酬率大都具有高狹峰與厚尾的現象,由於常態分配無法捕捉到厚尾的分佈情況。故本研究使用具有偏斜的分配與傳統對稱分配來做比較。實證結果發現在使用偏斜的分配模型,風險值的估計較傳統對稱分配更為精準。在波動性模型使用上,非厚尾95%與90%信賴水準之下以GARCH模型表現為佳,厚尾99.5%與99%信賴水準之下又以EWMA模型表現為佳。此外,風險值(VaR)並無法捕捉到尾端風險的損失,故本研究亦納入預期損失(Expected Shortfall),藉由預期損失提供較為完整尾端風險的資訊。可知常態分配,在預測ES的結果為較佳;而在波動模型當中,則以GARCH模型在預測ES的結果為較佳。
Recently, studies find that most financial Return of Asset have the leptokurtosis and the fat-tails phenomenon, because normal distribution can not capture the fat-tails. Therefore, this study compares the skewed distribution and the traditional symmetric distribution. The empirical research finds that value at risk (VaR) estimate using the skewed distribution is more accurate than that using the traditional symmetric distribution. Using the volatility model, the performance of confidence level at 95% and 90% is better under the GARCH model; the performance of confidence level at 99.5% and 99% is better under the EWMA model. In addition, value at risk can not capture the tail risk, so this study also included the expected shortfall (ES), which offered more complete extreme loss events and tail risk information. When predicting ES, using the normal distribution is better. Among the volatility models, the GARCH model predicts better ES.