Crude oil price forecasting based on internet concern using an extreme learning machine
Introduction
International crude oil price prediction has become an increasingly important issue. Crude oil plays a significant role in the global economy (Uri, 1996), with the crude oil market accounting for nearly two-thirds of the world’s energy demand (Alvarez-Ramirez, Soriano, Cisneros, & Suarez, 2003). A leap in the price of crude oil would result in inflationary pressure and an economic recession within most countries, and therefore would have a significant impact on the global economy. In contrast, a rapid decline in crude oil prices would prohibit economic development in oil-producing countries, thus creating political instability and potentially social unrest. As a result, it is necessary to investigate the inherent mechanisms of oil price fluctuations in order to reduce the potential risks of oil price volatility.
In general, crude oil prices are determined by supply and demand Hagen (1994), Stevens (1995), although they are also influenced by speculation and extreme events, which can intensify the price volatility and market instability. Numerous studies Demirer and Kutan (2010), Kaiser and Yu (2010), Zhang et al. (2009) have argued that additional market factors should be intercalated into an analytical framework for the prediction of crude oil prices. Oil prices are sensitive to oil-related events such as war, extreme weather, OPEC production stipulations, etc. Ji and Guo (2015) investigated the effects of four types of oil-related events on world oil prices and concluded that oil prices respond differently to specific events. Hamilton (2009) highlighted the fact that the Iraq War and the Iranian Revolution resulted in oil supply disruptions that of course impacted oil prices.
The recently introduced concept of internet concern (IC) involves the use of crude oil market search data for quantifying investor speculation. IC is now an important factor in explorations of the impact and magnitude of market concerns. Numerous studies have suggested that information extracted from the internet can contribute to the prediction of financial data Bollen et al. (2011), Bordino et al. (2012). Health economists have used Google search queries to provide early indicators in uncovering disease incidence and prevalence rates, while others have used such queries to predict consumer spending. For example, Choi and Varian (2012) highlighted the fact that queries data can be useful for indicating consumers’ planned purchasing. Other studies (e.g. Askitas and Zimmermann (2009), Bank et al. (2011), Preis et al. (2013)) have used Google search data to measure IC in financial markets. Li, Ma, Wang, and Zhang (2015b) used a Google search volume index to quantify investor attention and investigated the relationships between search data, differential trader positions and crude oil prices. Park, Lee, and Song (2016) also utilized internet search data from Google Trends for forecasting the short-term flow of Japanese tourists to South Korea. Yao and Zhang (2017) explored the effects and predictive power of the Google Index on crude oil prices by incorporating the Google Index into ARIMA and ARMA-GARCH models as an exogenous variable.
Various traditional statistical and econometric models, such as cointegration, GARCH, vector autoregression (VAR) and Markov models, have been adopted for analyzing oil markets Allegret et al. (2015), Maghyereh (2006), Salisu and Oloko (2015), Zhang and Wang (2015). Considering the nonlinear patterns and irregularities hidden within the oil price series, artificial intelligence models such as neural networks (NN; see Yu, Wang, & Lai, 2008), support vector machines (SVM; see Xie, Yu, Xu, & Wang, 2006), and genetic algorithms (GA; see Motlaghi, Jalali, & Ahmadabadi, 2008), have also been used for forecasting crude oil prices. For instance, Chiroma, Abdulkareem, and Herawan (2015) proposed a hybrid approach based on a genetic algorithm and neural network (GA-NN) for predicting the West Texas Intermediate (WTI) crude oil price. Baruník and Malinska (2016) proposed the use of a generalized regression framework based on neural networks for forecasting oil prices. Single layer feed-forward networks (SLFN) are used widely in classification and regression analysis applications. However, SLFN and other such gradient descent learning methods are time-consuming and suffer significant critical errors, such as over-fitting, local minima, etc. Huang et al. (2004), Huang et al. (2006) proposed a learning algorithm known as ‘extreme learning machine’ (ELM), which has performed well for predicting non-linear time series and has a better generalization performance than the gradient-based learning methods. ELM has been implemented widely for short-term wind power forecasting (Abdoos, 2016), hyperspectral imagery classification (Li, Chen, Su, & Du, 2015a), electricity price forecasting (Xiao et al., 2016), and online sequential prediction (Wang & Han, 2015).
In parallel to the development of forecasting technologies, numerous decomposition and construction methods have been developed, such as wavelet analysis (Shahbaz, Tiwari, & Tahir, 2015), singular spectral analysis (SSA; see Fenghua, Jihong, Zhifang, & Xu, 2014), and empirical mode decomposition (EMD; see Huang et al., 1998). EMD has been used as an effective analysis model in economics and finance He et al. (2016), Yu et al. (2008), Zhang et al. (2009). As an extension of EMD into the two-dimensional space, bivariate empirical mode decomposition (BEMD) was first proposed by Rilling, Flandrin, Gonalves, and Lilly (2007). BEMD simultaneously models the joint oscillating modes at each intrinsic mode function (IMF) and provides a robust estimate of asymmetry for nonlinear and nonstationary data Molla et al. (2011), Yang et al. (2011).
This paper analyzes internet attention on the crude oil market, along with the impacts of two oil-related events, namely abnormal climate incidents and war. Three IC indices for capturing the influences of internet attention are constructed and an ELM-based forecasting model is established that incorporates intrinsic modes and IC. The objectives are to:
- 1.
quantify the influence of emergencies on the crude oil market using the information extracted from the internet;
- 2.
examine the advantages of the BEMD-based modeling framework for analyzing the transmission between each IC index and oil price volatility, while characterizing the magnitude and dynamics of the impacts at various frequencies; and
- 3.
investigate the power of forecasting models with the aid of intrinsic modes and internet data for short-run crude oil price volatility.
Section 2 presents the methodology formulation of the basic BEMD and ELM, while Section 3 presents the main contribution, namely an IC analysis of the crude oil market based on BEMD and price forecasting using the ELM model. Section 4 reports experimental results, and Section 5 provides some conclusions.
Section snippets
Bivariate empirical mode decomposition
EMD is a signal processing technique that decomposes a univariate (real-valued) signal into waveforms by extracting all of the oscillatory modes embedded within the signal. The waveforms extracted by EMD are named intrinsic mode functions (IMF); these are modulated in both amplitude and frequency.
BEMD is a generalized extension of the EMD for complex signals, and is particularly suitable for estimating amplitude information simultaneously across different frequencies for two nonlinear and
Internet concern analysis and price volatility forecasting
Traders in commodity futures markets fall into two categories: physical traders and investors (speculators) of index funds. Physical traders, whose companies use oil in production, buy oil for future delivery, then sell at a fixed price. Since they know that the price of oil will change, they try to make predictions in order to minimize the risk, also known as hedging. In contrast, investors (or speculators) in commodity index funds are increasingly including commodity futures in their
Empirical analysis
We now investigate the performance of the ELM model under different schemes.
Conclusions
We have quantified the impact of internet attention on crude oil markets using a novel BEMD-based ELM modeling framework. The proposed modeling framework and experimental study provide a better analysis and forecasting of the crude oil market using easily accessible internet data. Specifically, our main contributions are as follows.
-
We use a new perspective, internet concern indices, to quantify the influence of emergencies on the crude oil market. We demonstrate that a growing
Acknowledgments
This work was supported by the Youth Innovation Promotion Association, CAS (No. 2014004), the National Center for Mathematics and Interdisciplinary Sciences (NCMIS) (No. 629092ZZ1), CAS, and the National Natural Science Foundation of China (NSFC Nos. 71771208 and 71271202).
Jue Wang. Professor of University of Chinese Academy of Sciences and associate editor of the Journal of Systems Science and Information. Research interests include computational intelligence, decision analysis and economic forecasting. She has published 70 papers and 6 books in IGI Global publisher and the Scientific Publisher in China. Principal Investigator of over 15 National Natural Science Fund projects, knowledge innovation projects of Chinese Academy of Sciences, and financial risk
References (44)
A new intelligent method based on combination of VMD and ELM for short term wind power forecasting
Neurocomputing
(2016)- et al.
Oil price shocks and global imbalances: Lessons from a model with trade and financial interdependencies
Economic Modelling
(2015) - et al.
Symmetry/anti-symmetry phase transitions in crude oil markets
Physica A. Statistical Mechanics and its Applications
(2003) - et al.
Forecasting the term structure of crude oil futures prices with neural networks
Applied Energy
(2016) - et al.
Twitter mood predicts the stock market
Journal of Computational Science
(2011) - et al.
Evolutionary neural network model for West Texas Intermediate crude oil price prediction
Applied Energy
(2015) - et al.
The behavior of crude oil spot and futures prices around OPEC and SPR announcements: An event study perspective
Energy Economics
(2010) - et al.
Stock price prediction based on SSA and SVM
Procedia Computer Science
(2014) - et al.
Extreme learning machine: theory and applications
Neurocomputing
(2006) - et al.
Oil price volatility and oil-related events: An internet concern study perspective
Applied Energy
(2015)
The impact of Hurricanes Gustav and Ike on offshore oil and gas production in the Gulf of Mexico
Applied Energy
How does Google search affect trader positions and crude oil prices?
Economic Modelling
An expert system design for a crude oil distillation column with the neural networks model and the process optimization using genetic algorithm framework
Expert Systems with Applications
Modeling oil price-US stock nexus: A VARMA-BEKK-AGARCH approach
Energy Economics
The determination of oil prices 1945–1995: A diagrammatic interpretation
Energy Policy
Crude-oil price volatility and agricultural employment in the USA
Applied Energy
Improved extreme learning machine for multivariate time series online sequential prediction
Engineering Applications of Artificial Intelligence
Bivariate empirical mode decomposition and its contribution to wind turbine condition monitoring
Journal of Sound and Vibration
Forecasting crude oil prices with the Google Index
Energy Procedia
Forecasting crude oil price with an EMD-based neural network ensemble learning paradigm
Energy Economics
Exploring the WTI crude oil price bubble process using the Markov regime switching model
Physica A. Statistical Mechanics and its Applications
Estimating the impact of extreme events on crude oil price: An EMD-based event analysis method
Energy Economics
Cited by (106)
A novel hybrid model for crude oil price forecasting based on MEEMD and Mix-KELM
2024, Expert Systems with ApplicationsFrom vineyard to table: Uncovering wine quality for sales management through machine learning
2024, Journal of Business ResearchNonferrous metal price forecasting based on signal decomposition and ensemble learning
2024, Journal of Process ControlA new hybrid deep learning model for monthly oil prices forecasting
2023, Energy Economics
Jue Wang. Professor of University of Chinese Academy of Sciences and associate editor of the Journal of Systems Science and Information. Research interests include computational intelligence, decision analysis and economic forecasting. She has published 70 papers and 6 books in IGI Global publisher and the Scientific Publisher in China. Principal Investigator of over 15 National Natural Science Fund projects, knowledge innovation projects of Chinese Academy of Sciences, and financial risk management projects of China’s commercial bank.
George Athanasopoulos. Associate Professor in the Department of Econometrics and Business Statistics, Monash University, Australia. His research interests include multivariate time series analysis, forecasting, non-linear time series, wealth and tourism economics. He is on the Editorial Board of the Journal of Travel Research. He has published in journals such as the International Journal of Forecasting, Journal of Econometrics, Journal of Business and Economics Statistics, Journal of Applied Econometrics, Computational Statistics and Data Analysis, Econometric Reviews, Applied Economics, Journal of Time Series Analysis, and Tourism Management.
Rob J. Hyndman. Professor of Statistics in the Department of Econometrics and Business Statistics, Monash University, Australia, and Editor-in-Chief of the International Journal of Forecasting. His research interests include business analytics, machine learning, forecasting, demography, computational statistics, and time series. He has held academic positions at Monash University, the University of Melbourne, Australian National University and Colorado State University. He is currently a director of the International Institute of Forecasters and an elected member of the International Statistical Institute.
Shouyang Wang. Professor of Academy of Mathematics and Systems Sciences of CAS, School of Economics and Management at University of Chinese Academy of Sciences and Editor-in-Chief of the Journal of Systems Science and Information. He is associate editor of 12 journals and Principal Investigator of 30 National Natural Science Fund projects. He has published 28 books and over 220 journal papers. His current research interests include economic forecasting, financial engineering, supply chain management and decision support systems.