本研究以捕捉機率密度的風險值模型為研究對象,我們應用Knight,Satchell and Tran (1995)的Double-gamma分配來計算風險值,並比較EWMA、混合常態分配、t分配、高斯核心密度函數、指數核心密度函數及極值分配的GEV與GPD分配等共八種模型的績效。實證的對象是臺灣83個店頭市場公司股票、等權投資組合及模擬投資組合。績效評比是以二項分配、條件涵蓋法與分配預測模型進行檢定。實證的結果顯示,在估計單日個股風險值方面,EWMA的表現最佳;在估計多日個股風險值方面,以Student's t分配的表現較為出色;而本文首度應用的Double-gamma分配在處理模擬投資組合風險值時其績效最優異。
This paper evaluates different probability density functions in the application of Value-at-Risk measures. We apply Knight, Satchell and Tran's (1995) Double-gamma distribution to measure VaR and examine models including EWMA, Mixture Normal, Student's t, Gaussian Kernel, Epanechnikov Kernel and Extreme Value distributions. The data include 83 Taiwan OTC stocks, equal weighted portfolio and simulated portfolio. The tests based on binomial test, conditional coverage and distribution forecasts show that the EWMA performs best for one-day holding period and the Student's t model tends to outperform the others for five-day holding period. In addition, the Double-gamma model also performs well for simulated portfolio.