School of Management, Huazhong University of Science & Technology
The National Natural Science Foundation of China (General Program, Key Program, Major Research Plan)
原油价格受国际政治、经济、军事、外交以及其他复杂因素的影响，这些因素的频繁变化使油价表现出随机波动，给原油投资及交易决策带来困难，因此，准确预测油价已成为能源领域学术界的研究热点。但是现有关于原油价格预测的文献大多数是预测原油价格的数值而不是变化方向，而且不是同时预测原油价格和波动率，因此无法给投资者充分的决策指导信息。为了填补这一研究空白，本文提出了一种结合转移网络(transition network，TN)、链接预测(link prediction，LP)、长短期记忆模型(long short-term memory，LSTM)和支持向量机(support vector machine，SVM)的新的混合模型TN-LP-LSTM-SVM来更精确地预测WTI期货次日价格变化方向和波动率大小，为投资者、能源相关企业和参与政策决定的政府人员提供有益的建议。本文比较了在不同的时间窗口下 (h∈[1,50]且h∈Z+) TN-LP-LSTM-SVM与CNN-SVM、LSTM和SVM的预测精度，发现在进行中长期预测时(h≥5)，TN-LP-LSTM-SVM总是稳健地优于CNN-SVM、LSTM和SVM。
Crude oil prices are influenced by international political, economic, military, diplomatic and other complex factors, and the frequent changes in these factors cause oil prices to exhibit random fluctuations, making crude oil investment and trading decisions difficult. Therefore, predicting oil prices accurately has become a hot research topic in the academic field of energy. However, most of the existing literature on crude oil price forecasting predicts the value of crude oil prices rather than the change direction, and does not predict crude oil prices and volatility simultaneously, thus can’t give investors sufficient information to guide their decisions. To fill this research gap, this paper proposes a new hybrid TN-LP-LSTM-SVM model combining transition network (TN), link prediction (LP), long short-term memory model (LSTM) and support vector machine (SVM) to predict the next-day price change direction and volatility size of WTI futures more accurately, providing useful advice for investors, energy-related companies, and government personnel involved in policy decisions. In this paper, we compare the prediction accuracy of TN-LP-LSTM-SVM with CNN-SVM, LSTM and SVM for different time windows (h ∈ [1, 50] and h ∈ Z+) and find that TN-LP-LSTM-SVM always outperforms CNN-SVM, LSTM and SVM robustly for medium and long term predictions (h ≥ 5).