A sentiment-based approach to predict energy price volatility using distilRoBERTa and GARCH models

Previous studies have extensively examined the impact of information on short-term energy price fluctuations, using various forms to extract sentiment, such as search volume and news headlines. However, the influence of social media data on energy prices has received little attention. Therefore, we extend the existing literature by using tweets to analyze the impact of social media on the change in energy prices. Furthermore, we propose a new approach to classify text data using the distilRoBERTa fill-mask task, which provides direct predictions of classification keywords, rather than manually categorizing them as the traditional classification task does. The sentiment volatility then shows a significant impact on the volatility of the crude oil and natural gas prices, although an asymmetric effect is only observed for WTI crude oil. Our findings also indicate that the exponential GARCH model offers the best fit for energy price returns and sentiment volatility. In general, incorporating sentiment volatility enhances the performance of modeling the short-term volatility of crude oil and natural gas prices and suggests that social media seem to impact the uncertainty level and the expectation of customers and investors regarding energy prices.