Research On International Crude Oil Price Prediction Based On Stream Learning

Crude oil is one of the most important energy resources on earth,currently accounting for nearly one-third of the global energy consumption.International crude oil prices have a big impact on the global environment and economy.Industries,governments and individuals pay great attention to oil prices.Thus,forecasting international crude oil prices has been a popular research topic in both academia and industry.Although many methods have been developed for predicting oil prices,it remains one of the most challenging forecasting problems due to the high volatility of oil prices.In this dissertation,we propose a new approach for international crude oil price prediction based on stream learning.The oil price prediction model based on stream learning is continuously updated whenever new oil price data are available.Thus,the prediction model evolves over time and can better capture the changing pattern of international crude oil prices.Our stream learning approach not only overcomes the disadvantage of traditional econometric approaches,which cannot capture the nonlinearity of international crude oil prices,but also overcomes the drawback of traditional machine learning approaches,which cannot effectively predict non-stationary international crude oil prices.Therefore,the new approach can make more accurate predictions on future international crude oil prices.In addition,our stream learning approach can effectively handle continuous international crude oil price data,even if these data are generated with high speed.For each new data point,it only requires small constant time and fixed amount of memory to update the prediction model,and there is no need to re-train the model using all the old and new data points.This dissertation applies stream learning techniques to the field of international crude oil price prediction,which is novel,both domestically and internationally.This novel approach is strongly supported by experimental results.To evaluate the forecasting ability of our streaming learning model,we compare it with the no-change model,the artificial neural networks model and a cutting-edge forecast combinationmodel.We apply a multi-horizon quantitative analysis method.The forecast time horizons include 1-month,3-month,6-month,9-month and 12-month.We use two standard performance metrics: Mean Squared Prediction Error(MSPE)and Directional Accuracy Ratio(DAR).The experiment results show that for both the West Texas Intermediate crude oil spot price and the U.S.refiner acquisition cost for crude oil imports from January 1992 to September 2012,the stream learning model achieves the highest accuracy in terms of both MSPE and DAR(lowest MSPE,highest DAR)over most time horizons(1-month,6-month,9-month and 12-month).This dissertation proposes a new powerful computational and analytical tool to the field of energy forecasting.