過去應用GARCH模型於期貨避險方面的研究大多僅侷限於雙變量,多變量避險的研究仍相對較少有學者涉足,主要原因為維度上的詛咒,當維度太高即會造成參數過多難以收斂的問題。本文試圖採用結合獨立成分分析法(Independent Component Analysis,簡稱ICA)的ICA-GARCH模型,透過其維度降階與觀測波動叢聚的特點,探討該模型於高維度下的期貨避險績效。本研究以澳幣、日幣、英鎊、瑞士法郎與加拿大幣等五種貨幣的期貨現貨為樣本建構投資組合。實證結果顯示,ICA-GARCH模型於雙變量的績效不比BEKK-GARCH模型差,而在雙變量以上的投資組合避險下,其維度降階的優點更加明顯,各期的樣本外績效均明顯優於其他模型。
Previous research on GARCH futures hedging is mainly focused on bivariate. This is because the problem of curse of dimension inherent in the multivariate portfolio hedging that creates problems of over parameterization and convergence. This article attempts to apply Independent Component Analysis GARCH (ICA-GARCH) for futures hedging. ICA-GARCH possesses properties of dimension reduction and volatility clustering observed frequently in the financial data and avoid the problems of curse of dimension. Five foreign exchange data, Australian dollar, Japanese Yen, British pound, Swiss Francs and Canadian dollars are investigated here. Empirical results show that in general ICA-GARCH is not inferior to BEKK-GARCH model and for the case of portfolio hedging with more than one commodity, the benefits of dimension reduction is getting clearly and ICA-GARCH consistently outperforms BEKK-GARCH out-of sample.