Can Economic Policy Uncertainty, Volume, Transaction Activity and Twitter Predict Bitcoin? Evidence from Time-Varying Granger Causality Tests

Abstract

We examine the predictive power of economic policy uncertainty, volume, transaction activity, and Twitter on Bitcoin between 27 December 2013 and 11 February 2019 using the recently proposed time-varying Granger causality tests of Shi et al. (2018). First, of particular interest, we show that volume can only predict Bitcoin returns during two episodes (August 2016-January 2017 and May 2017-June 2017) based on a Wald test with a recursive evolving procedure under a homoskedasticity error assumption. However, volume cannot predict volatility under any specifications. Secondly, both US economic policy uncertainty and equity market uncertainty indices, which are used as proxies for policy uncertainty, have no effect on predicting Bitcoin returns. Thirdly, transaction activity also cannot predict Bitcoin returns. Lastly, the number of tweets about Bitcoin can Granger cause the volume of Bitcoin (for example, March 2015-August 2015 and January 2016-February 2019) but not returns or volatility.