Forecasting Oil Price Trends With News Sentiment

The growth or decline in oil prices will have a significant economic impact on the whole world. Under such background, efficient and accurate predictions for crude oil price are critical for a stable economic development. Despite such attempts, oil price prediction has remained a difficult problem due to its complexity and irregularity. The price of oil is basically determined by balancing the amount of oil the net oil-exporting countries can supply with the demands of the net importing countries, but the irregularity is caused more by the shocks on the supply-side, which may be political disputes or sudden changes in external economic factors. In such cases, a precise prediction of the values of the oil price will be difficult to obtain. However, a rough prediction of the upward and downward changes of the price can still be helpful for decision making. With the rapid development of the Internet and big data technologies, a variety of news data to be processed continues to witness a quick increase. In fact, content of news information represents the real-time evaluation on future trend of financial markets. The mainstream news and information will be reflected in the emotional final of the investment behavior of financial practitioners, and thus affect the stock market. If we can take advantage of this big data on the Internet, we can analyze and forecast the movement direction in crude oil prices. Generally speaking, this paper introduce news sentiment, which is extracted based on a dictionary-based approach, into crude oil price trend forecasting and employ several powerful machine algorithms (i.e., SVM, DT, LogR and BP) to verify the predictive power of news sentiment. Also, we use a Granger causality analysis to investigate whether news sentiment correlate with changes in crude oil price and to determine the predictive lag order. For illustration and verification purposes, the crude oil future price in West Texas Intermediate (WTI) market and the news of crude oil market are collected from the international news agency Thomson Reuters are taken as sample data. Empirical results of Granger causality analysis statistically show that, news sentiment has significant Granger causality relation with crude oil price and the variations of news sentiment more likely lead to changes in crude oil price, which indicate the predictive power of news sentiment. Empirical results of trend prediction show that adding news sentiment have significant effect on prediction accuracy compared to using only historical crude oil price values.