Title: A VAR-SVM model for crude oil price forecasting

Authors: Lutao Zhao; Lei Cheng; Yongtao Wan; Hao Zhang; Zhigang Zhang

Addresses: School of Mathematics and Physics, Science and Technology University of Beijing, Beijing 100083, China; Centre for Energy and Environmental Policy, Beijing Institute of Technology, Beijing 100081, China ' Economic and Information Research Branch, China Coal Research Institute, Beijing 100031, China ' School of Mathematics and Physics, Science and Technology University of Beijing, Beijing 100083, China ' Centre for Energy and Environmental Policy, Beijing Institute of Technology, Beijing 100081, China ' School of Mathematics and Physics, Science and Technology University of Beijing, Beijing 100083, China

Abstract: In recent years, the complexity and variability of international crude oil price have had an increasingly greater impact on society's economic development. Therefore, the accurate forecasting of crude oil price is helpful to maintain economic stability and avoid risks. This paper analyses the influencing factors, including market factors and non-market factors and then uses the Vector Autoregression (VAR) model to measure the relationship between oil price and those factors. Based on the results of VAR model, we put forward a new model - VAR-SVM, which is based on VAR and Support Vector Machine (SVM). Using VAR-SVM, we can make more accurate prediction of crude oil prices. Genetic Algorithm (GA) is employed to select model parameters. From the empirical results we can see that VAR-SVM model is superiority in accuracy and effectiveness comparing with the other forecasting models, such as VAR model, CGARCH model, Artificial Neural Network (ANN) model and autoregression SVM model.

Keywords: crude oil prices; price forecasting; VAR-SVM model; GAs; genetic algorithms; vector autoregression; VAR; support vector machines; SVM; modelling.

DOI: 10.1504/IJGEI.2015.069485

International Journal of Global Energy Issues, 2015 Vol.38 No.1/2/3, pp.126 - 144

Received: 31 Jul 2014
Accepted: 11 Nov 2014
Published online: 18 May 2015 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article