透過您的圖書館登入
IP:3.129.247.196
  • 學位論文

台灣實質有效匯率預測-倒傳遞類神經網路分析之應用

Applying Backpropagation Nerual Network to Predict an Real Effective Exchange Rate in Taiwan

指導教授 : 黃志祥
若您是本文的作者,可授權文章由華藝線上圖書館中協助推廣。

摘要


台灣為一對外貿易依存度很高的小型開放經濟體系,實質有效匯率除了能夠明確表達一國貨幣在匯率上相對於其他主要貿易往來國家的變動程度,更能夠看出該國在國際中是否具有競爭力的指標。本研究分兩個研究設計來探討,研究設計一,以歷史資料,針對台灣實質有效匯率進行預測;研究設計二,以經濟基本變數來作預測,藉由七個經濟基本變數:國內外相對生產力差異、國外淨資產比例、國內外相對貿易條件、政府消費比例、國內消費比例、國內外相對經濟開放程度、國際原油價格,來對台灣實質有效匯率進行預測。 近年來,被廣為研究與討論的類神經網路資訊處理系統,因其具有網路穩定性高、良好的錯誤容忍度、歸納推廣的能力以及非線性特質等優點,因此被廣泛應用在各個領域。而在財務金融方面,以類神經網路為基礎的預測方法,已有許多成功案例。本文將以倒傳遞類神經的五種演算法進行預測。 本文實證結果,以歷史資料來作預測時,倒傳遞類神經的LM演算法預測台灣實質有效匯率的方向預測正確性效果最佳,且樣本內為選取十二個月訓練期間,是對台灣實質有效匯率的預測能力為最佳的選取範圍。 以經濟基本變數來作預測,則是以倒傳遞類神經的共軛梯度演算法的預測能力最佳。各變數對台灣實質有效匯率間之重要性與相關程度最佳的為國際原油價格;重要性與相關程度最小的為國外淨資產比例。本研究希望藉由研究結果,作為投資者進行投資及企業從事貿易之參考,來減低因匯率波動所產生之風險。

並列摘要


Taiwan is a small open economy with a high dependence on trade. The real effective exchange rate is able to express the degree of fluctuation in the exchange rate relative to the other major trade countries, and more can be seen in the country whether in international competitiveness indicators. The study consists of two studies designed to investigate, study design one to historical data for forecasting Taiwan's real effective exchange rate; study design two to basic economic variables to make predictions, with seven basic economic variables: relative productivity differences, net foreign assets, the relative domestic trade, government consumption, domestic consumption, domestic relative degree of economic openness, international crude oil prices, to predict the real effective exchange rate in Taiwan. In this study, we use five algorithms of backpropagation nerual network to predict real effective exchange rate. Recently, neural networks are a field of research which has enjoyed a rapid expansion and great popularity in both the academic and industrial research communities. Form the statistician’s point of view they are analogous to nonparametric, nonlinear regression models. This is particularly useful in financial engineering applications, where much is assumed and little is known about the nature of the processes determining asset prices. The empirical result, use historical data to predict real effective exchange rate in Taiwan, the Levenberg-Marquardt algorithms of backpropagation neural network is best approach for Directional Change and twelve months is the best training rang. Use basic economic variables to predict real effective exchange rate in Taiwan, the conjugate gradient algorithm of backpropagation neural network is best approach. Importance and relevance of the best is the international crude oil prices; importance and relevance of the smallest is the net foreign assets. This research work supposes that the result of empirical analysis can be regard as a refrence for investing and trading in order to eliminate risks resulting form fluctuated exchange rate.

參考文獻


[2]曹添旺、賴景昌、鍾俊文,郭炳伸與蔡文禎,2002,“新台幣實質有效匯率之動態分析”,台灣經濟預測與政策,32,頁93-130。
[15]柏婉貞,2010,“油價可以解釋實質匯率走勢嗎?亞洲新興國家之驗證”,東吳經紀商學學報,69期,頁29-46,6月。
[16]呂佳芹,2009,“應用時間序列、演化式類神經網路與灰預測方法在匯率預測績效之比較”,朝陽科技大學財務金融系碩士論文。
[12]萬怡如,2012,“實質匯率預測力-以台灣為例”,國立中正大學國際經濟學博士學位論文。
[14]邱至中,2003,“長短期匯率預測模式績效之比較”,國立成功大學財務金融研究所,碩士論文。

延伸閱讀