Tags:Cryptocurrencies, dynamic model averaging, predictability and time-varying parameter regressions
Abstract:
Despite the interest in cryptocurrencies has recently increased, in the literature a consensus has not yet been reached on whether they represent or not a new asset class, spanning risks and payoffs sufficiently different from the traditional ones. We contribute to this debate by studying the exposure of cryptocurrency returns to stock market risk factors (namely, the six Fama French factors), to precious metal commodity returns, and to cryptocurrency-specific risk-factors (namely, crypto-momentum, a sentiment index based on Google searches, and supply factors, i.e., electricity and computer power). Because economic facts lead us to believe that those exposures are likely to be time-varying, we rely on Bayesian methods, which incorporate dynamic model averaging. These methods not only feature time-varying coefficients, but also allow for the entire forecasting model to change over time. We estimate our flexible models on weekly data for four popular cryptocurrencies, namely Bitcoin, Ethereum, Litecoin and Ripple. We find that cryptocurrencies are not systematically exposed to stock market factors, precious metal commodities or supply factors with exception of some occasional spikes of the coefficients during our sample. On the contrary, they display a time-varying but significant exposure to a sentiment index and to crypto-momentum. However, cryptocurrencies display considerable diversification power in a portfolio perspective and as such they can generate differential Sharpe ratios and certainty equivalent returns in spite their overall predictability turns out to be weaker vs. traditional asset classes.
Dissecting Time-Varying Risk Exposures of Cryptocurrency Markets