Group transfer entropy with an application to cryptocurrencies

The detection of informational leadership is a core issue in financial market microstructure. We use effective group transfer entropy (EGTE) as a measure for the predictability of a stochastic process using lagged observations on multiple related processes within the same system. We propose an appropriate bootstrap to derive confidence bounds and show by means of a simulation study that standard linear approaches in economics and finance, such as vector autoregressions and Granger-causality tests, are not well suited to detect information transfer. We empirically examine the markets for cryptocurrencies using intraday data and reveal that the dependencies are mostly of nonlinear nature, highlighting the applicability of EGTE in the context of this new financial product.