應用自組織映射神經網路於外幣市場匯率動態評估系統設定

Abstract

近年來人工智慧技術領域被廣泛應用在財金工程上，促使新的研究能跳脫了過去傳統的統計模型限制，尤其是類神經網路被廣泛的應用在財務預測與行為模是探討上，在各種類神經網路之中，自組織映射圖神經網路(Self-Organizing Map；SOM)對於物件分類與行為預測，有著相當良好的參考性，同時又可以大量處理多種非線性的時間序列變數，將多維度的評估指標同時納入同一個模型處理，不僅可以幫助研究者更貼近現實交易市場現況，透過資料前處理將能量趨勢術質輸入後，更可藉此衡量出研究目標未來的走向趨勢。 本研究系統設計利用自組織映射圖神經網路結合總體經濟基要指標，進行貨幣匯率在外幣市場上的族群分類與績效驗證，並進一步提出「貨幣匯率靜態體質檢定分析模型」以及「貨幣匯率動態體質檢定分析模型」，將多維度的總體經濟基要指標經資料前處理後，輸入本研究提出的匯率模型，得到視覺化二維平面的靜態分群圖與動態的趨勢分群圖，並藉由貨幣在分群圖上的相對位置，進行匯率行為軌跡分析，提出一個投資模擬系統進行模型的驗證，希望藉此驗證匯率行為與基本面之間的關係與匯率行為的趨勢能量。 實驗結果顯示，依據本研究設定之投資模擬系統對SOM分群圖的貨幣匯率體質軌跡變化進行分析投資，可獲得超額利潤，證明本研究提出的「貨幣匯率體質檢定投資模型」及其模擬系統設定，可有效藉由該國總體經濟基要，對外幣市場的貨幣匯率靜態與動態趨勢進行分群，能提供投資人以視覺化的方式，直接對貨幣匯率靜態與動態體質軌跡變化進行分析及投資。本研究證實預測模型之預測準確率顯著地優於隨機漫步模式。

In recent years the field of artificial intelligence technology is widely used in financial engineering, lots of new studies show that financial studies are no longer restricted by traditional financial engineering model that are constructed by statistic. Form the views of mentioned studies; artificial neural network (ANN) could improve the performance of a new methodology both in the object classification and the ability of financial forecasting. The SOM (Self-Organizing Map；SOM), one of ANN family, can only process accurately a large number of nonlinear time series variables but also project multi-dimensional targets into a two-dimensional map to define different clusters. This map resembles a landscape in which it is possible to identify borders that define different clusters. Consequently, SOM can help the new methodology to study closer to the reality of market status and can also reveal the trends of the behavior of foreign exchange rate according suitable data input. In this paper, we analyze whether the behavior of foreign exchange rate can be predicate according to proposed SOM model that is base on the macroeconomic variables from 1990 to 2006. Furthermore, we adopt 11 macroeconomic variables of 22 countries and applied an concept of trends to visualize and compress multi-dimensions data for investors’ decision .To do this, we establish a system according to the dynamic analysis model of foreign exchange rate. We employ a two-level portfolio selection model, which is base on SOM (proposed system) and buy-and-hold (a popular strategy). We find that the hedge strategy guided by proposed system (SOM) may reap excessive profits than buy-and-hold strategy. Consequently, we find that proposed system is suitable for the dynamic evaluation of foreign exchange rate.