Nonlinear dependence in cryptocurrency markets

Abstract We are interested in describing the returns and volatility dynamics of major cryptocurrencies. Very high volatility, large abrupt price swings, and apparent long memory in volatility are documented features of such assets. We estimate a multivariate stochastic volatility model with discontinuous jumps to mean returns and volatility. This formulation allows us to extract a time-varying shared average volatility and to account for possible large outliers. Nine cryptocurrencies with roughly three years of daily price observations are considered in the sample. Our results point to two high volatility periods in 2017 and early 2018. Qualitatively, the permanent volatility component seems driven by major market developments, as well as the level of popular interest in cryptocurrencies. Transitory mean jumps become larger and more frequent starting from early 2017, further suggesting shifts in cryptocurrencies return dynamics. Calibrated simulation exercises suggest the long memory dependence features of cryptocurrencies are well reproduced by stationary models with jump components.