Crude Oil Price Forecasting With Online News Sentiment Analysis

Crude oil is not only an important strategic resource,but also one of the important commodities in the global financial market.Its price volatility has a far-reaching impact on the global economy.The prediction of oil price is very important for decision makers.Its value is immeasurable.Because there are numerous and complex factors affecting the oil price,the oil price series show complexity and non-stationarity.Therefore,it is very difficult to predict oil prices,which often leads to prediction errors.In this context,the main purpose of this paper is to explore ways to improve the performance of oil price prediction and reduce the prediction error.Online news contains a large amount of valuable information about events and public opinion factors,which is helpful for forecasting research.Sentiment analysis technology can be used to mine and quantify information.It becomes an effective means of mining online news.Therefore,this paper mainly studies the oil price forecasting with online news sentiment.From the micro-,meso-and macro-levels,that is,the sentiment of news paragraphs,news sentiment categorized by topics and news summarization sentiment.The specific research contents are as follows:(1)Crude oil price forecasting with online news paragraphs sentimentBased on the existing literature,some research hypotheses are put forward.Sentiment scores of online news articles and paragraphs are incorporated to several oil price prediction models respectively.The improvement of prediction results is compared and analyzed.According to the final empirical results,the hypotheses are tested and the conclusion is drawn.Oil price forecasting models with news sentiment have better daily forecasting performance,especially fine-grained news sentiment,i.e.news paragraphs sentiment.At the same time,compared with low-frequency oil price data,high-frequency oil price data with news sentiment can achieve better prediction results.(2)Crude oil price forecasting with online news sentiment classified by news topicsTopic models are applied to identify the topics of online news and transform the news into topic distribution data.News articles sentiment is classified by topics.Topic distribution and news sentiment under each topic are incorporated to the oil price prediction models respectively.The results show that,compared with benchmark models,the oil price forecasting models with the online news topic features and the sentiment features under the topic classification have good performance in oil price forecasting research.(3)Crude oil price forecasting with online news text summarization sentimentUsing text automatic abstracting technology,the online news is abstracted.And using sentiment analysis technology,the summarization sentiment is calculated.Then incorporating the news summarization sentiment into the crude oil price prediction models.The forecasting results show that news summarization sentiment has a significant impact on the daily oil price forecasting.Empirical research results show that compared with the benchmark models,the three forecasting models with micro-,meso-and macro-levels news sentiment proposed in this paper have better prediction performance.