一籃子貨幣避險—效率前緣與類神經網路之比較

CURRENCY BASKET HEDGE — THE COMPARISON BETWEEN EFFICENT FRONTIER AND ARTIFICIAL NEURAL NETWORK

吳欣芸(Hsin-Yun Wu)

指導教授 ： 賀蘭芝

共同指導教授 ： 鍾彩焱(Tsai-Yen Chung)

元智大學/管理學院/管理研究所/碩士(2008年)

摘要

壽險公司的海外投資部位，讓壽險公司享受到較高的投資收益，然而為規避匯率風險以及大舉換匯造成新台幣對美金市場價格的波動，央行多要求壽險公司應進行100%避險後才得以將資金匯出海外，因此除了匯率風險外，避險成本的波動將成為海外投資收益高低另一個因素。2006年底時，海外投資避險後收益一度曾因避險本高達4%而低於國內投資收益，因此為了達到規避匯率風險以及避險成本的降低兩個目的，一籃子貨幣組合避險策略因應而生。

本研究以「限制追踨誤差」(Constant Tracking Error)概念為模型基礎，分別建置兩個模型來各求導出一組一籃子貨幣組合的最適權重，再比較這兩個模型之最適權重的避險效果以及避險成本是否降低，兩模型的資料假設分別使用各貨幣的歷史日收盤匯率以及各貨幣在未來欲避險期間的日收盤匯率(以類神經網路的倒傳遞網路預測而得)，前者是的資料採用是假設過去歷史走勢與未來相同；後者則是假設倒傳遞網路所預測與未來相同。在倒傳遞網路的工具選用方面，為提高不具程式撰寫基礎使用者亦能輕易入門，本硑究採用Matlab之套裝工具列— NNTool，讓複雜的倒傳遞網路概念能更簡單明瞭。

實證結果顯示，兩模型皆可達到匯率風險規避以及避險成本降低的目的，但兩模型在匯率風險規避以及避險成本降低各有其優勢，投資組合管理者可視其重視匯率風險的規避，亦或是避險成本的降低，而選擇不同的模型來進行一籃子貨幣避險組合的建置。

關鍵字

外匯避險 ； 一籃子貨幣組合 ； 效率前緣

並列摘要

Life insurance companies in Taiwan enjoy better returns for their investment abroad. These investment in foreign currency are therefore subject to exchange risk and substantial FX volatility when insurances move their massive currency position from one to another, thus it is required by the Central Bank that these overseas positions are 100% FX-hedged. Another main cause for fluctuating returns is the swinging hedging cost by which total return for foreign investment in late 2006 was effectively overtaken by domestic one. Currency Basket Hedge was then developed for FX-risk-aversion and the pursuit of lower hedging cost.

Based on the model of Constant Tracking Error, this research aims to identify, by using two models constructed, the optimal weighting of each composite in a currency basket. Results from respective models are then compared for effectiveness in hedging and cost level. The data sources include historical daily closing price and forward closing price within a given time frame (by using the Back-Propagation Network to forecast). The former assumes the historical price movement is identical to the one for the future, whereas the latter assumes what is forecast by the Back-Propagation Network is the same vis-à-vis the future price movement. Also, this research uses NNTool within the Matlab Package, as it is more suitable for users who are not familiar with computer programming, which will also enable the concepts of Back-Propagation Network to better comprehended.

This research shows that both models serve the objectives reasonably well, yet they have different advantages in the two objectives and portfolio managers can choose accordingly for the set-up of currency basket hedging.

並列關鍵字

basket currecny hedge ； efficient frontier

參考文獻

1.張宗載 (2006)，「一籃子貨幣避險」，碩士論文，台灣大學財務金融研究所。

2.王怡雯 (2006)，「一籃子貨幣之資產配置與風險衡量」，碩士論文，台灣大學財務金融研究所。

3.葉柏村 (2002)，「運用類神經網路預測匯率—以歐元為例」，碩士論文，中原大學企業管理研究所。

1.G. Peter Zhang（2004）,Neral Networks in Business forecasting, 1st edition, Idea Group Publishing, U.S.A.

2.J. Wesley Hines（1997）,Fuzzy and Neural Approaches in Engineering, 1st edition, John Wiley & Sons, Inc, U.S.A., 52-117