The role of textual analysis in oil futures price forecasting based on machine learning approach

Abstract This paper offers an innovative approach to capture the trend of oil futures prices based on the text‐based news. By adopting natural language processing techniques, the text features obtained from online oil news catch more hidden information, improving the forecasting accuracy of oil futures prices. We find that the textual features are complementary in improving forecasting performance, both for LightGBM and benchmark models. Besides, event studies verify the asymmetric impact of positive and negative emotional shocks on oil futures prices. The generated text‐based news features robustly reduce forecasting errors, and the reduction can be maximized by incorporating all features.