應用多重類神經網路於基本面與技術面因素預測新台幣兌美元匯率

Abstract

採用浮動匯率制度已是全球外匯市場之必然趨勢，如何有效地掌握匯率波動之趨勢將成為一個很重要的課題。自傳統之總體經濟理論至現今的技術分析方法，已有許多學者及業界專家不斷地探討影響匯率之因素，並且嘗試對匯率的變動做預測。近年來，人工智慧方法學在財務金融領域上之應用也蓬勃發展，其中類神經網路被公認為適合解決類似之非線性問題。因此本研究使用多重類神經網路之架構，並結合總體經濟基本面和技術面之輸入指標，提出一個具有匯率預測能力之人工智慧模式。

本文研究1989年至2007年新台幣兌美元日資料，結合物價、利率、貨幣供給、進出口以及生產力等五個總體經濟基本面因素，以及實務界較常使用之七個技術指標，經過資料前處理後，分別當作三個類神經子網路之輸入變數，並將模擬結果彙整於主網路做總和評判。實驗結果顯示，多重類神經網路對於匯率預測之效果顯著優於隨機漫步模式，並且也顯著優於使用單一類神經網路之預測結果。

It is a inevitable trend that more and more countries adopt the floating exchange rate regime, so how to anticipate the fluctuation of exchange rate is a very important task. Since the traditional model of exchange rate and technical analysis method is provided, there are many scholars and specialists trying to research the factors which make exchange rate change. In the present, Neural Network is a novel methodology artificial intelligence which is widely applied to solve complicated financial problems. According to these backgrounds, a model with Multi-Neural Network integrating fundamental and technical analysis is provided.

In this study, the original data is the exchange rate between USD and NT from 1989 to 2007. Combining the five macro economics factors of price level, interest rates, money supply, imports/exports, and productivity and the seven technical indicators in practice, the data is pre-processed as input variables of the three sub networks. The results of simulation are integrated into the master network for total evaluation. The result of this experiment shows that Multi-Neural Network is significantly more effective than random walk model and single neural network model.