波動度為金融商品於一段時間其價格變化累積的觀察指標，其在財務工程領域中有著無與倫比的學術地位，諸多其衍生相關之學術研究領域，舉凡由資產定價模型回推估算的隱含波動度、由高頻資料價格變動累積的真實波動度…等。
本文的研究目標以元大台灣50 ETF之日頻率真實波動度預測作為研究標的，利用Engle et al. (2008) 所提出整合相異頻率資料訊息的GARCH-MIDAS方法，整合由主成分分析 (PCA) 萃取過後之低頻總體經濟因子，並參酌Song et al. (2023) 所使用之流動性相關變數，再搭配極限梯度提升 (XGBOOST) 建構預測模型，以五項殘差相關指標 (MAE、MSE、RMSE、SMAPE、RMSPE) 衡量模型成效，最後以Shapley (1951) 所提出的Shapley value 回推機器學習模型的預測邏輯，增強此模型的解釋性。
實證結果顯示，GARCH-MIDAS 方法所取得之短期波動度有著低估且無法有效追蹤短期真實波動度型態的問題，但其於長期型態上有著不俗的追蹤能力，故將其整合每日流動性相關變數，並輔以機器學習與貝葉斯超參數 (Bayesian Optimization for Hyperparameter Tuning) 修正能達到很好的預測與短期型態追蹤效果，並於Shapley value 模型解釋時，GARCH-MIDAS (Generalized Autoregressive Conditional Heteroskedasticity - Mixed Data Sampling)之短期因子有著非常重要的變數邊際貢獻。

Volatility is an indicator of the accumulated price changes of a financial product over a period of time, which has an unparalleled academic status in financial engineering. There are numerous related academic research fields derived from it, such as implied volatility inferred from asset pricing models, realized volatility accumulated from high-frequency price changes, and so on.
The research objective of this paper is to forecast the daily realized volatility of the Yuanta/P-shares Taiwan Top 50 ETF as the research subject. We employ the GARCH-MIDAS model proposed by Engle et al. (2008) to integrate information from mixed-frequency data, incorporating the low-frequency macroeconomic factors extracted by principal component analysis (PCA) and considering the liquidity-related variables used by Song et al. (2023). We then construct a forecasting model using extreme gradient boosting (XGBoost) and evaluate the model performance using five residual-related metrics (MAE, MSE, RMSE, SMAPE, RMSPE). Finally, we use the Shapley value proposed by Shapley (1951) to explain the prediction logic of the machine learning model, enhancing its interpretability.
The empirical results show that the short-term volatility obtained by the GARCH-MIDAS model has problems of underestimation and inability to effectively track short-term realized volatility patterns. However, it has decent tracking ability for long-term patterns. By integrating daily liquidity-related variables and correcting with machine learning and Bayesian optimization for hyperparameter tuning, we can achieve good forecasting and short-term pattern tracking performance. In the Shapley value model explanation, the short-term factor of GARCH-MIDAS (Generalized Autoregressive Conditional Heteroskedasticity - Mixed Data Sampling) has a very important variable marginal contribution.

陳慶賢 (2018)。使用 GARCH-MIDAS 模型對台灣股市波動度進行建模和預測。應用經濟研究, 11(2), 115-130。
王昱成, 吳宗杰 (2017)。GARCH-MIDAS 模型在台灣股市波動度建模中的應用：實證研究。風險與金融管理學報, 10(4), 43。
Alaali, F. (2020). Forecasting stock market volatility using GARCH, value-at-risk and expectedshortfall models: Evidence from Saudi Arabia. Quarterly Review of Economics and Finance, 78, 27-43.
Asgharian, H., Hou, A. J., & Javed, F. (2013). The importance of the macroeconomic variables in forecasting stock return variance: a GARCH‐MIDAS approach. Journal of Forecasting, 32(7), 600-612.
Bollerslev, T. (1986). Generalized autoregressive conditional heteroskedasticity. Journal of Econometrics, 31(3), 307-327.
Chen, N. F. (2018). Stock volatility forecasting and model selection. Review of Quantitative Finance and Accounting, 50(1), 185-221.
Dong, J. (2022). Modelling and forecasting long-term volatility with Markov-switching asymmetric jump GARCH–MIDAS models. Journal of Forecasting, 41(6), 988-1012.
Engle, R. F., Ghysels, E., & Sohn, B. (2008). On the economic sources of stock market volatility. NYU Working Paper No. FIN-08-046.
Ghysels, E., Qian, H., & Zhang, X. (2019). Machine learning methods for financial applications. Available at SSRN 3475723.
Lundberg, S. M., & Lee, S. I. (2017). A unified approach to interpreting model predictions. Advances in Neural Information Processing Systems, 30.
Shapley, L. S. (1953). A value for n-person games. Contributions to the Theory of Games, 2(28), 307-317.
Song, Y., Tang, X., Wang, H., & Ma, Z. (2023). Volatility forecasting for stock market incorporating macroeconomic variables based on GARCH-MIDAS and deep learning models. Journal of Forecasting, 42(3), 243-267.
Taha, A. A., & Malebary, S. J. (2020). An intelligent approach to credit card fraud detection using an optimized light gradient boosting machine. IEEE Access, 8, 79184-79200.
Tong, E., Lim, C., Zhu, A., Cheng, D., & Wolters, J. (2020). A hybrid GARCH-MIDAS approach to modeling and forecasting stock market volatility. The Journal of Risk Finance.
Zhu, Y., & Shao, X. (2017). Stock volatility forecasting using GARCH-MIDAS-XGBoost model. Finance Research Letters, 22, 38-42.
Zhou, X., Pan, W., & Xiao, Y. (2021). Stock market volatility forecasting with XGBoost-Embedded Residual Neural Networks. Expert Systems with Applications, 178, 114962.