Oil price volatility and new evidence from news and Twitter

In this paper, we develop semantic-based sentiment indices through relevant news and Twitter feeds for oil market using a state-of-the-art natural language processing technique. We investigate the predictability of crude oil price volatility using the novel sentiment indices through a hybrid structure consisting of generalized autoregressive conditional heteroskedasticity and bidirectional long short-term memory models. Findings show that media sentiment considerably enhances forecasting quality and the proposed framework outperforms existing benchmark models. More importantly, we compare the predictive power of news stories with Twitter feeds and document the superiority of the news sentiment index over the counterpart. This is an important contribution as this paper is the first study that compares the impact of regular press with that of social media, as an alternative informative medium, on oil market dynamics. •Oil price volatility predictability is checked using news and Twitter sentiments.•BERT, GARCH, and BiLSTM models are used.•Findings reveal media sentiment enhances forecasting accuracy.•News sentiment outperforms Twitter sentiment in forecasting oil price volatility.•The weight of evidence is of more predictive power than its strength for oil price volatility.