Assessing causal relationships between cryptocurrencies and investor attention: New results from transfer entropy methodology

Abstract

Studies apply non-parametric wavelet Granger causality testing to investigate bi-directional causalities of cryptocurrencies with Twitter and Google. However, this method only provides the existence of information flows without quantization and assumes time series are linear. Considering this, we highlight transfer entropy as an alternative, model-free methodology. We quantify the impact of search-engine attention (Google Trends) and social-media attention (Twitter) on cryptocurrency returns, employing in turn Shannon and Rényi transfer entropy methodologies. We document levels of bi-directional causalities, showing that tail events are more informative than center observations in the cryptocurrency market.