GARCH-MIDAS 對波動性與涉險值預測的應用

Abstract

本文應用 Fang et al. (2020, Journal of Empirical Finance) 所提出結合懲罰項之自我迴歸條件異質變異數-混合頻率模型 (GARCH-MIDAS)，進行股票與債券市場報酬率之波動度與涉險值（VaR）的日頻率預測。在實證過程中考慮了多個月頻率的預測變數；此外，也加入傳染病風險變數做為新的解釋變數。實證結果表示所考慮的實證設定在股票市場波動度與涉險值預測均具有相對的優勢；另外，就債券市場的波動度與涉險值預測而言，雖整體的預測能力下降，但本文的實證設定尚能展現一定的預測能力。

This thesis applies the generalized autoregressive conditional heteroskedasticity mixed data sampling (GARCH-MIDAS) with penalized framework proposed by Fang et al. (2020, Journal of Empirical Finance) to generate daily forecasts of volatility and Value at Risk (VaR) for stock and bond index returns. Beyond daily returns, we incorporate several monthly macroeconomic and financial predictors, and we further augment the model with indicators that capture infectious disease risk. Empirically, our specification outperforms both the traditional GJR-GARCH model and the quarterly-frequency GARCH-MIDAS framework. While its overall predictive power declines for bond index returns, the proposed specification still delivers a meaningful level of accuracy in that market.