高頻率資料通常呈現豐富的資料特性,包括波動叢聚、槓桿效應、波動持續性,及厚尾、高峰、偏態等非常態分佈特性。雖然GARCH模型在厚尾分佈的設定方式,於實證研究上被發現可處理財務金融市場上常見波動聚集和厚尾現象;但對於上述其他重要的資料特性,則無法妥適的加以涵括。本文針對文獻上的不足,提出較為一般化的GJRI GARCH-EGB2模型,其不只考量厚尾、高峰、及偏態等資料分佈特性,並進一步將波動不對稱及一致性納入模型考量。應用到亞洲八個主要國家的股市報酬率資料,發現不論樣本內或樣本外的檢測,本文所提出的方法皆優於傳統的模型設定方式。此外,相較於一般文獻對於波動不對稱性之重視,本研究發現針對分佈假設的妥適設定,對於模型效能的提昇更為顯著。
High frequency stock return data tend to exhibit characteristics such as volatility clustering, volatility persistence, leverage effects, and properties of nonnormal unconditional distributions reflected in the form of skewness, high peakedness, and excess kurtosis. Although traditional GARCH models that employ leptokurtic distributions have been found useful to account for the conditional heteroscedasticity and leptokurtosis, they have difficulty in accommodating other stylized effects commonly observed in high frequency data. This paper attempts to rectify this deficiency by introducing a more general GJR IGARCH-EGB2 model, which not only considers the flexible distributional characteristics associated with the exponential beta distribution, but also incorporates the asymmetric conditional variance and integrated GARCH process into model consideration. Likelihood ratio tests , goodness of fit tests, distribution plots, and out-of-sample forecasts generate a preponderance of evidence to support the innovative GJR IGARCH-EGB2 specification over conventional competing alternatives presented in the literature.