期刊
ENERGY ECONOMICS
卷 71, 期 -, 页码 114-127出版社
ELSEVIER SCIENCE BV
DOI: 10.1016/j.eneco.2018.02.004
关键词
Crude oil price; Combination forecasting; PPM; BMA; TVIP-MRS; WP-STSM
类别
资金
- National Science Foundation of China [71473155]
- Young Star of Science and Technology Plan Project of China's Shaanxi Province [2016KJXX-14]
- Free Control Programs of Basic Research Funding of Xidian University [JB160603]
In view of the importance and complexity of international crude oil price, this paper proposes a novel combination forecast approach that captures a variety of fluctuation features in crude oil data series, including change points, regime-switching, time-varying determinants, trend decomposition of high-frequency sequences, and the possible nonlinearity of model setting. First, product partition model-K-means (PPM-KM) model is used to detect change points in the oil price sequence. Next, we apply a time-varying transition probability Markov regime switching (TVTP-MRS) model to identify the regime-switching characteristic. Then, we use Bayesian model averaging (BMA) to filtrate main determinants at each regime. Finally, the time-varying parameter structure time series model (TVP-STSM) is used to decompose the oil sequence, capture the time-variation of coefficients in volatile upward regime, and forecast the crude oil price. Compared with some other competing models and benchmark model of ARIMA, the newly proposed method shows superior forecasting ability in four statistical tests. Besides, we make scenario prediction on WTI crude oil price to examine the implementation effect of OPEC cut-off agreement at the end of 2016. OPEC production and U.S. shale oil production are used as two scenario variables, and the WTI price is forecasted fluctuating around 50 dollar/barrel based on three scenario prediction. We conclude that WTI crude oil price would take a shock upstream tendency in the short-term but the rising scope would not be large. (C) 2018 Elsevier B.V. All rights reserved.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据