再探金融指數已實現波動度預測

Abstract

波動度預測在金融領域是重要的研究方向之一，過去有許多學者提出不同的模型來提高預測的準確性。Corsi (2009)提出異質自我回歸(Heterogeneous Autoregressive, HAR)模型，使用過去一天、一周、一個月(1、5、22交易日)的已實現日內波動度來預測未來波動度，僅使用十分簡單的結構就擊敗其他更複雜且具長期記憶型的計量模型。而本研究以HAR模型為基礎，採用能夠捕捉更多日內價格變動且受市場結構雜訊影響較小的日內波動度計算方法，試圖提高預測精確度。同時，探討原始HAR模型的參數設定在具有不同交易性質的商品上是否合理，並通過擴展參數來尋找最佳的參數組合。實證結果顯示，在特定情況下，HAR模型的設定確實具有優異表現。然而，隨著商品交易性質和投資人行為的不同，參數選擇的傾向也會有所不同，並且發現某些特定參數在跨商品上能達到一致的優良結果。

Volatility prediction is a crucial research direction in the financial field, with many scholars having proposed various models to enhance prediction accuracy. Corsi (2009) introduced the heterogeneous autoregressive (HAR) model, which uses realized volatility over the past day, week, and month (1, 5, and 22 trading days) to forecast future volatility. Despite of its simple structure, the HAR model outperformed other more complex econometric models with long-term memory. This study builds on the HAR model, employing an intraday volatility calculation method that captures more intraday price movements and is less affected by market structural noise, aiming to improve prediction accuracy. Additionally, it explores whether the parameter settings of the original HAR model are reasonable for commodities with different trading characteristics and extends the parameters to find the optimal combination. Empirical results indicate that, in specific scenarios, the HAR model settings indeed exhibit excellent performance. However, with varying trading characteristics of commodities and investor behavior, the preference for parameter selection also differs. Furthermore, it is found that certain parameters achieve consistently excellent results across different commodities.