Comparison of risk forecasts for cryptocurrencies: A focus on Range Value at Risk

We forecast the Range Value at Risk (RVaR) of main cryptocurrencies using the GARCH model with different error distributions. We compare the performance of the different forecasts using a score function. The normal and asymmetric normal distributions presented the best performance for RVaR. Our findings suggest that the main driver for the RVaR of cryptocurrencies is the conditional standard deviation and not the distribution of the stochastic term. For the Value at Risk (VaR) and Expected Shortfall (ES), non-normal distributions present the best performance. We also note the advantages of RVaR over ES regarding regulatory arbitrage and model misspecification. •We forecast the risk of main cryptocurrencies using Range Value at Risk (RVaR).•Normal distribution presents the best performance to predict the RVaR.•RVaR is less sensitive than ES to regulatory arbitrage and model misspecification.