近年來資產波動估計的重點圍繞在使用高頻交易資料，但如何將具有高資訊價值的已實現波動融入GARCH模型中以預測未來資產的波動仍是一個熱門的議題。本研究主要以Hansen et al. (2011)所提出的已實現GARCH模型為基礎，延伸出四個議題進行研究探討。第一個議題使用點波動預測與分位預測評估使用以報酬率或是以變幅為基礎的已實現波動搭配已實現GARCH 模型，何者可以達到較優異的預測績效?第二個議題探討使用該模型進行多期波動預測時，是否較按比例(scaling up)增加的方式準確?第三個議題為考慮價格跳躍的影響，採用免除跳躍干擾的已實現波動融入模型是否能提升波動預測績效?第四個議題將此模型搭配DCC-GARCH模型，應用到多變量波動預測相關的財務應用，如投資組合風險值、最小變異避險與波動擇時策略等。上述四個議題預期能對於該模型在波動預測的文獻上有所貢獻。
Recent studies about volatility estimation for asset returns focus on estimating with high-frequency data, and how to incorporate highly informative realized volatility with GARCH model to forecast asset volatilities remains a quite hot issue. This study proposes four issues based on the realized GARCH model by Hansen et al. (2011). The first issue is to use alternative realized measures of volatility, such as return-based or range-based realized volatility, in realized GARCH to achieve superior performance for point and quantile forecasts. The second issue focuses on whether multi-horizon forecasts of the volatility by the realized GARCH model are more accurate than scaling up method for point and quantile forecasts. The third issue is to examine whether using jump-robust realized measure improve one- and multi-period-ahead volatility forecasts. The fourth issue extends univariate volatility forecasts to multivariate volatility financial applications, such as portfolio VaR, minimum variance hedge and volatility timing strategy. Hopefully, these four issues are expected to contribute in literature related to volatility forecasting.
