應用類神經網路預測高級銅極之價格及其影響因素之分析

The Application of Artificial Neural Networks to Forecast The Price of Refined Copper

賴光榮(Kuang-Jung Lai)

指導教授 ： 田方治

國立臺北科技大學/管理學院/工業工程與管理系所/碩士(2006年)

https://doi.org/10.6841/NTUT.2006.00059

摘要

本研究應用類神經網路，以國內物價指數及其他可能影響高級銅極價格之因素，預測其漲跌趨勢與影響因素關係之分析，其目的有三：
1、期以更準確及迅速的判斷，預測國際高級銅極之交易價格；
2、將所預測之數據及早提出因應措施，以降低企業之採購成本；
3、分析國內之物價指數因素，對預測國際高級銅極價格之影響。
本研究以過去十一年，可能影響國際高級銅極價格，或與其有連動關係的四個構面，共計二十一項相關因素之歷史資料為研究主體，應用倒傳遞類神經網路 (Back Propagation Neural Network, BPNN)與資料包絡分析法(Data Envelopment Analysis, DEA)，將其分成三個群組作為配適的輸入項目，其目標在選出誤差績效最理想之網路模式予以預測銅價之短期走勢，並分析不同影響因素，對預測國際高級銅極價格之影響程度。
本研究結果顯示國內物價指數對銅價之預測之貢獻度不高，確切影響銅價走勢之項目有黃金、白金、鈀金、白銀、原油與高級銅極之交易價格及LME銅的庫存量與台幣、英鎊對美元匯率等九項，以此九項預測銅價短期之走勢最可能區間為3.3至3.9US/LB。故本研究建議在此混沌不明，基金積極介入之際，銅價短期仍持續波動，但至3.9US/LB時，有一賣壓，企業應小心評估，慎防追高。

關鍵字

類神經網路 ； 預測 ； 配適 ； 高級銅極 ； 資料包絡分析 ； 誤差績效

並列摘要

The prediction of precious metal has been an important issue in domestic and international lately. This research applies Artificial Neural Networks to predict the refine copper’s transaction price using the domestic price index and other possibility factors. The objective of this research includes:
1.to predict the transaction price of the international refine copper rapidly and accurately;
2.to make a quick react, either buy in or sell out, based on the derived predicted price in order to reduce the purchase cost;
3.to analyze the influence of the corresponding domestic price index factors to predict international refine copper’s transaction price.
After collecting corresponding data related the international refine copper’s transaction price in the past 11 years, this research decompose the 21 relevant factors into four main domains of factors. Three different groups of backpropagation neural networks (BPNN) are constructed to conduct the learning and predicting based on the collected data, and to use the data envelopment analysis approach to be elected the best BPNN’s construction of the error performance. The goal of the BPNN’s is to predict the short-term tendency of the copper’s price, and analysis the influence degree of predicting the copper’s price with different influence factors.
After learning, testing, and validation of BP networks, the experimental results show that the domestic price index has less degree of influence to the prediction of the copper's price; but nine factors, including the trade prices of gold, platinum, palladium, silver, crude oil and copper, and volume of stock of the copper, and exchange rate of Taiwan dollar, Pound sterling to the exchange rate of U.S. dollar, are highly correlated to the price tendency of copper. By using the nine factors, we predict that the transaction price of copper will be fallen into 3.3 and 3.9US/LB in the short period of future.

並列關鍵字

Artificial Neural Network ； Forecast ； Fitting ； Refine copper ； Data Envelopment Analysis ； Error Performance

參考文獻

[18] 高強、黃旭男、Sueyoshi，管理績效評估，台北：華泰文化事業股份有限公司，2003，第3-24頁。

[17] 李茂興譯，作業研究，台北：揚智文化事業股份有限公司，2000，第587-623頁。

[26] 陳國玄，人工神經網路與統計方法應用於台灣上市電子類股價指數預測與分類之研究，碩士論文，國立成功大學統計學研究所，台南，2004。

[24] 李惠妍，類神經網路與迴歸模式在台股指數期貨預測之研究，碩士論文，國立成功大學管理學院，台南，2003。

[2] A.K. Jain, J. Mao, and K.M. Mohiuddin, "Artificial neural networks: a tutorial. Comput." IEEE March, 1996, pp. 33–44.