This research explores the weekly crude oil price data from U.S. Energy Information Administration over the time period 2009 - 2017 to test the forecasting accuracy by comparing time series models such as simple exponential smoothing (SES), moving average (MA), and autoregressive integrated moving average (ARIMA) against machine learning support vector regression (SVR) models. The main purpose of this research is to determine which model provides the best forecasting results for crude oil prices in light of the importance of crude oil price forecasting and its implications to the economy. While SVR is often considered the best forecasting model in the main stream literature, this research investigates its computational insights in terms of parameter selections and overfitting potential, in addition to exploring forecasting accuracy and model comparison. The results of this research can be generalized to forecast other business and economic time series data such as stock market prices, product sales, and government statistics.
He, Xin James
"Crude Oil Prices Forecasting: Time Series vs. SVR Models,"
Journal of International Technology and Information Management: Vol. 27
, Article 2.
Available at: https://scholarworks.lib.csusb.edu/jitim/vol27/iss2/2
Business Intelligence Commons, Communication Technology and New Media Commons, Computer and Systems Architecture Commons, Data Storage Systems Commons, Digital Communications and Networking Commons, E-Commerce Commons, Information Literacy Commons, Management Information Systems Commons, Management Sciences and Quantitative Methods Commons, Operational Research Commons, Science and Technology Studies Commons, Social Media Commons, Technology and Innovation Commons