Explaining Cryptocurrency Price Trends: Statistical Analysis of Social Media Posts vs Market Prices

Yajing Zhi, Lanfang Kong, Samed Atouati, T.-H. Hubert Chan, Mauro Sozio

Abstract: We conduct a statistical analysis investigating whether one could explain the cryptocurrency price trends by their historical market prices or social media posts. Our study focuses on Bitcoin while considering both volume and sentiment of Bitcoin-related posts on Reddit. We developed a classifier based on BERT, which is well known to perform well for domain-specific tasks. We then study the Granger causality link between either post volume, post sentiment and Bitcoin price. Finally, we evaluate the returns of simple investment strategies based on either post volume, post sentiment or historical Bitcoin prices. Our most important findings, based on almost five years of data from Reddit, are that there are causality links between post volume and Bitcoin price, as well as post sentiment and Bitcoin price that are statistically significant, with the former one being more pronounced. Moreover, our study suggests that post volume on Reddit better explains Bitcoin price trends than historical price trends themselves. 1