Does Social Media Sentiment Predict Bitcoin Trading Volume?

Jayit Saha, Smit Patel, Frank Xing, Erik Cambria

Abstract: Description Social media sentiment is proven to be an important feature in financial forecasting. While the effect of sentiment is complex and time-varying for traditional financial assets, its role in cryptocurrency markets is unclear. This research explores the predictive power of public sentiment on Bitcoin trading volume. We develop a novel sentiment analysis pipeline for processing Bitcoin-related tweets and achieve state-of-the-art accuracy on a benchmark dataset. Our pipeline also leverages information gain theory to incorporate the impact of textual and non-textual features. We use such features to discern a non-linear relationship between public sentiment and Bitcoin trading volume and discover the optimal predictive horizon for Bitcoin. This research provides a useful module and a foundation for future studies and understanding of Bitcoin market dynamics, and its interaction with social media buzzing.