應用灰關聯分析與類神經網路於預測人民幣匯率之研究

The Exploratory Study on the Prediction of the RMB Exchange Rate by Using Gray Relational Analysis and Neural Network

鄭漴瑋(Tsung-Wei Cheng)

指導教授 ： 李樑堅

共同指導教授 ： 黃永成(Yung-Cheng Huang)

義守大學/管理學院/財務金融學系/碩士(2014年)

https://doi.org/10.6343/ISU.2014.00461

摘要

人民幣從三十年前國際貨幣體系中的默默無聞，到今天越來越多國家的貨幣兌換地點、ATM，乃至中央銀行儲備庫開始使用人民幣，人民幣走過了一條不尋常的國際化道路，也直接反映了中國大陸經濟在三十多年間的崛起速度。另外也有愈來愈多學者針對人民幣匯率做過相關研究，如楊錦文(2013)利用倒傳遞類神經網路來針對長短期人民幣匯率做預測；另外，覃凡丁、奉欽亮 (2012)也運用迴歸分析法來探討人民幣匯率漲跌和進出口總額及外匯儲備之間的關係。因此分析人民幣之漲跌趨勢有其探討的意義及研究價值所在。
本研究首先透過灰關聯分析篩選出貿易收支、國際收支、通貨膨脹、信貸、市場預期、外匯儲備、利差、GDP、投資率、物價水平、貨幣供給等十一項重要影響人民幣匯率變化因子。再利用2004 年至2012年的月資料建構倒傳遞類神經網路模型，以找出能夠預測未來人民幣匯率漲跌最佳的模型。而多元迴歸模型則是將11個因子輸入，並以逐步迴歸法保留較少自變數，分析與依變數的關係，建立更精簡有效的迴歸模型。最後以2013年的資料驗證，比較多元迴歸模型及倒傳遞類神經網路模型兩者預測人民幣匯率漲跌趨勢，以界定較佳模型，也希望能夠提供給投資人做為參考依據。
研究結果發現倒傳遞類神經網路模型在預測人民幣匯率漲跌成效優於多元迴歸模型，但多元迴歸模型相對穩定，誤差較小，且倒傳遞類神經網路模型預測人民幣匯率漲跌的正確率高達75％，對人民幣投資而言具有一定參考價值。

關鍵字

人民幣匯率 ； 灰關聯分析 ； 倒傳遞類神經網路 ； 多元迴歸

並列摘要

RMB becomes popular than thirty years ago in currency exchange, nowadays more and more ATM and reserve currency of the central bank in many countries start to make use of RMB. RMB goes through an unusual process in the internationalization, and it also directly reveals that the economy of People’s Public of China (PRC) rapidly spring up during the thirty years. More and more scholars and researchers have studied the field of RMB exchange rate. Yang（2013）took advantage of back propagation neural network to predict RMB exchange. In addition, Feng and Qin（2012）also made use of regression analysis to assay the relations between the fluctuation in RMB exchange rate and total imports-exports and reserve currency. Therefore, to explore and analyze the fluctuation trend in RMB is meaningful and valuable.
In this study, the first step is to screen out 11 important factors of the RMB exchange rate, including “balance of trade,” “international balance of payments,” “inflation,” “credit,” ”market expectation,” “reserve currency,” “interest margin,” “GDP,” “investment rates,” “price level” and “money supply.” Then it builds a back propagation neural network database, from 2004 to 2012, and tries to figure out the best model to predict the RMB exchange rate in the future. Secondly, it puts 11 factors from grey relational analysis into the model and uses stepwise regression procedure to build a simple and efficient multiple regression model. At last, according to the data of RMB exchange in 2013, the study tries to compare these two models in accuracy of prediction for the trend of exchange rate in RMB. And then, it hopes that the study can define and provide a better model to the investors as references.
This study result shoes that back propagation neural network is better than multiple regression model on the prediction of RMB exchange rate.However, multiple regression model is relatively more stable and fewer errors than back propagation neural network. In accuracy of the prediction, back propagation neural network is about 75% and thus it is a worthy reference for RMB investing.

並列關鍵字

RMB exchange ； grey relational analysis ； back propagation neural network ； multiple regression model

參考文獻

吳有龍、王天津、陳盈君、廖敏秀 (2007)，台幣對多國外幣匯率灰預測研究，2007南台灣資訊科技與應用研討會，(382–388頁)，高雄：美和技術學院。

黃永成 (2011)，結合灰關聯分析之模糊連續遺傳演算法對選擇權之評價，資訊管理學報，18(1)，133–153。

鄭美幸、詹志明 (2002)，灰色理論與時間序列模型在匯率預測績效上之比較，台灣金融財務季刊，3(2)，95–104。

張瓊文、張瑞芳 (2010)，應用基因演算法與到傳遞類神經網路於匯率預測模型之開發，嘉南學報，36，270–279。

高長、蔡慧美 (2003)，大陸外匯體制變遷及人民幣升值趨勢分析，經濟前瞻，90，64–69。