Authors: Konstantinos Gkillas, Maria Tantoula, Manolis Tzagarakis

Title: Bitcoin realized volatility and transaction activity

Abstract

We study the predictive value of transaction activity in the Bitcoin network for the realized volatility of bitcoin returns constructed by high-frequency data. As an alternative modeling approach to the popular linear heterogeneous autoregressive model, we provide out-of-sample forecasts for realized volatility of bitcoin returns by means of machine learning algorithms and in particular by Random Forests. With the use of additional covariates including measures of jumps, we find that on-blockchain transaction activity does improve the out-of-sample forecast accuracy at all the forecast horizons considered.