虛擬貨幣之波動度預測 - 多變量 GARCH-MIDAS 模型

Abstract

本研究旨在應用多變量 GARCH-MIDAS 模型，探討比特幣市場波動率的來源及其預測能力。傳統的 GARCH 模型雖然能有效捕捉短期波動，但在納入不同頻率資料（如每月總體經濟變數與每日加密貨幣報酬）時存在限制。GARCH-MIDAS 模型能夠整合不同頻率的資料，並區分短期與長期波動來源，提供更全面的波動率建模架構。

本研究納入共 21 個潛在解釋變數，涵蓋四大類型：一般波動指標（如實現波動度Realized Volatility與恐慌指數 VIX）、美國總體經濟變數（如美國 CPI、PPI 等）、不確定性相關指數（如 EPU、MPU、GRI），以及加密貨幣專屬變數（如 UCRY、VCRIX、CFGI）。資料期間涵蓋 2019 年 7 月至 2024 年 7 月，特意避開早期市場尚未成熟的階段。

研究核心在於驗證兩項假說：（1）加密貨幣專屬變數相較於傳統金融變數更能有效解釋與預測比特幣波動率；（2）結合多種潛在波動來源的模型預測能力優於單一來源模型。實證結果顯示，VIX、美國總體經濟指標與某些不確定性相關指數（如 GRI）在單變數模型中展現顯著解釋力，而加密貨幣專屬變數（如 UCRY、CFGI）雖也有顯著解釋力，但模型表現不佳，反而是在預測能力方面表現優異。最終，雙變數模型整合 VIX 與其他解釋變數後，顯著提升了模型整體預測準確度。因此，本研究指出考量加密貨幣市場的獨特特性與波動結構，運用多變量 GARCH-MIDAS 模型進行分層式波動來源建模，能更有效解釋並預測其波動率，對風險管理與資產配置具有實務意涵。

This thesis examines Bitcoin volatility using the multivariate GARCH-MIDAS model, which integrates variables sampled at different frequencies by separating short- and long-term volatility components. The study uses data from July 2019 to July 2024—a period when the cryptocurrency market became more mature—and incorporates 21 explanatory variables from four categories: traditional volatility measures (e.g., realized volatility, VIX), U.S. macroeconomic indicators, policy uncertainty indices, and crypto-specific variables (e.g., UCRY, CFGI).

Two hypotheses are tested: (1) crypto-specific variables offer better explanatory power than traditional financial indicators, and (2) models combining multiple sources of volatility outperform single-source models. Results show that while variables like VIX and U.S. CPI have strong explanatory power, crypto-specific indices such as UCRY and CFGI also contribute significantly to forecasting accuracy.

The study further demonstrates that two-variable GARCH-MIDAS models—especially those combining VIX with crypto-specific variables—outperform single-variable models in both in-sample and out-of-sample forecasts. These findings emphasize the need to consider both traditional and market-specific factors when modeling cryptocurrency volatility, offering practical insights for risk management and financial decision-making in digital asset markets.