石油期貨之價格預測 -倒傳遞類神經網路、Elman類神經網路 、回饋式模糊類神經網路之比較

The Prediction of Crude Oil Futures Prices - Comparison aming Backpropagation Neural Networks,Elman Recurrent Neural Networks and Recurrent Fuzzy Neural Networks.

林瑞文(Jui-Wen Lin)

指導教授 ： 胡為善

中原大學/商學院/企業管理研究所/碩士(2006年)

https://doi.org/10.6840/cycu200600455

摘要

近年來，由於中東戰事持續不斷以及恐怖活動發生的影響，全球處於不穩定中，造成國際油價變動相當劇烈，令各國政府與企業不得不尋求一個準確的油價預測方法，來避險或從中套利。
本研究使用了三種常用的預測模型來預測NYMEX的原油期貨價格，倒傳遞類神經網路(Backpropagation Neural Network)、Elman類神經網路( Elman Neural Network )、回饋式模糊類神經網路(Recurrent Fuzzy Neural Network)，並比較此三種類神經網路於財務預測的績效表現。
本研究針對NYMEX原油期貨日資料，以倒傳遞類神經網路、Elman類神經網路與回饋式模糊類神經網路來進行預測能力之比較。根據本論文的實證研究顯示，不論是使用倒傳遞類神經網路、Elman類神經網路與回饋式模糊類神經網路，或是使用不同的訓練筆數，MSE值均小於0.0026767。由此可知，以類神經網路來預測石油期貨價格是相當適合，預測績效亦相當不錯。
另外，在學習訓練的績效上，三種類神經網路具有一致性，均是訓練筆數越多，學習績效越好。
在比較三個模型的預測績效上，回饋式模糊類神經網路的平均MSE為最小、其次是亦為回饋式類神經網路的Elman類神經網路，最後則是前向式的倒傳遞類神經網路。因此，經由本研究的實證得知，回饋式類神經網路之預測能力優於前向式類神經網路。而且將模糊理論與回饋式類神經網路結合，能大幅提升預測的能力。

關鍵字

倒傳遞類神經網路 ； 類神經網路 ； 模糊回饋式類神經網路 ； Elman回饋式類神經網路

並列摘要

During the past three years, oil price has changed dramatically and terrorists’ attacks caused the turbulent uneasiness of the global economy. Consequently, governments and corporate managers around the world actively sought effective methods to forecast the oil price more accurately than before for the purposes of hedging and arbitraging.
The purpose of this study is to predict the crude oil futures prices more accurately than traditional methods by using three popular non-parametric methods, namely, Backpropagation Neutral Networks (BPNs), Elman Recurrent Neural Networks (ERNNs), and Recurrent Fuzzy Neral Networks (RFNNs). This work also compares the learning and predictive performance among BPNs, ERNNs and RFNNs, and explores how training time impacts predictive accuracy.
The results show that the use of these three non-parametric methods to forecast the crude oil futures prices was appropriate since their values of MSE were all less than 0.0026767. Additionally, the learning ability was consistent by employing different training times. This investigation also indicates that the more training times the networks took, the better learning performance the networks have under most circumstances, the only exceptional case occurs at part two under FRNN model, where MSE is slightly less than that obtained from part three.
Regarding the predictive power of the three artificial neural networks (ANNs), this study finds that RFNNs has the best predictive power and BPN has the least predictive power among the three ANNs. This investigation also confirms that the predictive power can be enhanced by combining Fuzzy theory with the Recurrent Neural Network.

並列關鍵字

Artificial Neural Networks (ANNs) ； Recurrent Fuzzy Neural Networks (RFNNs). ； Backpropagation Neural Networks (BPNs) ； Elman Recurrent Neural Networks (ERNNs)

參考文獻

許志義、洪育民 (1992)，「國際油價分析與預測」，中華經濟研究院出版。

張俊弘 (2004)，「歐元外匯選擇權之評價-倒傳遞類神經模型與回饋式類神經模型之比較」，中原大學企業管理研究所碩士論文。

葉柏村 (2002)，「運用類神經網路預測匯率-以歐元為例」，中原大學企業管理研究所碩士論文。

Allen Hobbs, Nikolaos G. Bourbakis (1995), “A NeuroFuzzy Arbitrage Simulator for Stock Investing”, Computational Intelligence for Financial Engineering.

Baba, N. and Kozaki, M. (1992), "An Intelligent Forecasting System of Stock Price Using Neural Networks, " IJCNN, 1, PP.371-377.