Due to the complex supply-demand structure and influences from many unpredictable events, crude oil market is very difficult to predict. Instead of relying only on historical market data, this paper proposes a novel approach to predict daily crude oil price and volatility using topics and sentiment features from related financial news-headlines. Specifically, this paper is an early attempt to extract comprehensive text features from online news, and utilize them in the forecasting of crude oil price and volatility. Ten different topics were identified from news sample using LDA model, and two sentimental indicators (polarity score and subjectivity score) were constructed from news-headlines. Both the topic features and the sentiment features were incorporated in price movements and volatility forecasting. Empirical results of five different time windows revealed that the text-based forecasting models obtain prior performance compared with benchmark models.
Li, Xuerong; Shang, Wei; and Wang, Shouyang, "Crude Oil Price Movement and Volatility Forecasting based on Online News" (2017). PACIS 2017 Proceedings. 164.