使用自我組織特徵映射圖於台灣外幣市場匯率趨勢預測之研究

Abstract

匯率波動對以國際貿易為主軸的台灣影響至鉅。其影響大者，足以是一家中型上市公司的整年獲利。對匯率趨勢的準確預測成了競爭激烈的跨國企業致勝的關鍵。 隨著資訊技術的進步，人工智慧在財務預測上的應用愈來愈廣泛。其中以類神經網路最為顯著。經過許多實證發現，類神經網路能大幅改善預測的準確度。其中的自我組織特徵映射圖(Self-Organizing Maps, SOM)網路具備將高維度的資料映射在較低維度的空間上的特性，及在圖形辨識(Pattern Recognition)上的能力。本研究採用自組織映射圖演算法進行匯率K線型態的學習，並進而預測匯率下一基期的走勢以產生實際的買賣建議訊號，提供投資人買賣操作依據。 本研究結果證實SOM 網路可以有效的對K 線型態進行分群，並對預測的K 線型態進行買賣訊號判讀，據以產生實際的投資建議，同時也發現SOM在買賣訊號判斷上仍會有過度僵化的情形，輔以倒傳遞類神經網路將有助於改善這個問題。

Fluctuation of foreign exchange rate had tremendous influences towards Taiwan’s market since Taiwan’s economy had highly business dependence on international trade. The loss due to the fluctuation of foreign exchange was so big that could reach the annual profit of a mid-sized, listed company in TSEC Market. Therefore, the ability to forecast exchange rate precisely had become the key to compete with other transnational companies. Due to the advanced information technology, Artificial Intelligence, especially Artificial Neural Network, applied more in financial forecast. Based on the experiment, Artificial Neural Network could improve the accuracy of financial forecast. The Self-Organizing Maps (SOM), one of the Artificial Neural Network algorithm, is a method to reduce the dimension of data and display the data in low dimension space, and capable with pattern recognition. In order to forecast the reliable trading signal of the next trend period for the investors, this research had adopted the Self-Organizing Maps algorithm for the K-Chart pattern learning of Taiwan foreign exchange rate. The conclusion of this research had demonstrated that SOM network is capable of grouping K-Chart pattern effectively. Based on the forecasted K-Chart pattern, the reliable trading signal could be determined and the suggestion of investment could be provided as well. However, since the trading signal determination provided by SOM was not dynamic enough, BPN could be the aid to improve the case.