- 學位論文
多變量GARCH模型對台灣利率預測能力之比較研究
A Comparison of Multivariate GARCH Forecasting Ability on Taiwanese Interest Rates
摘要
本文以台灣的長期利率(十年期公債殖利率)及短期利率(三十天期銀行承兌匯票月利率)的預測績效(均方根誤差)作為探討主題，討論幾個主要著名的 GARCH 模型，包括多變量CC、DCC、GJR、BEKK及ABEKK等GARCH與單變量GARCH、EGARCH 在台灣長、短期利率上的預測績效，輔以Z test作為檢定績效的依據，希望能藉由這些預測績效的評估，以供投資大眾作參考。 對長期利率來說，多變量模型之利率預測績效幾乎都優於單變量模型。由於大多數文獻較著重在樣本外的預測結果，本文因此以樣本外分析為主。實證結果顯示：(1) 在樣本外短期利率預測上我們發現，在0.1的顯著水準下，波動不對稱模型之預測績效皆顯著優於波動對稱模型、動態相關性多變量模型顯著優於相對應的單變量模型、動態相關性模型也全都顯著優於相對應的靜態相關性模型。(2) 在樣本外長期利率預測績效上，動態相關性模型都優於靜態相關性模型，但只有BEKK-GARCH是顯著的。
並列摘要
In this paper, we examine the forecasting performance (root mean square; RMSE) of several important GARCH models including univariate GARCH, EGARCH and multivariate CC, DCC, GJR, BEKK, and ABEKK-GARCH models for Taiwanese long-term and short-term interest rates. The forecasts are estimated with Z test, and in terms of statistic criteria, we hope these results could be good indicators for public investors. The empirical evidence confirms that both on in-sample and out-of-sample forecasting performance the multivariate models are approximately superior to the univariate models. Previous studies, however, concentrate more on the out-of-sample forecasts, and so does this article. Then we find that at the short-term forecasting horizons the asymmetric models outperform the symmetric models at the 0.1 significant level. Furthermore, we find that the dynamic correlation multivariate models significantly outperform the corresponding univariate models and the dynamic correlation models significantly outperform the corresponding static correlation models. At the long-term forecasting horizons, the dynamic correlation models outperform the static correlation models, but only BEKK-GARCH is significantly superior to the static correlation models.