A comparison of methods for forecasting value at risk and expected shortfall of cryptocurrencies

Abstract

Several procedures to forecast daily risk measures in cryptocurrency markets have been recently implemented in the literature. Among them, long-memory processes, procedures taking into account the presence of extreme observations, procedures that include more than a single regime, and quantile regression-based models have performed substantially better than standard methods in terms of forecasting risk measures. Those procedures are revisited in this paper, and their value at risk and expected shortfall forecasting performance are evaluated using recent Bitcoin and Ethereum data that include periods of turbulence due to the COVID-19 pandemic, the third halving of Bitcoin, and the Lexia class action. Additionally, in order to mitigate the influence of model misspecification and enhance the forecasting performance obtained by individual models, we evaluate the use of several forecast combining strategies. Our results, based on a comprehensive backtesting exercise, reveal that, for Bitcoin, there is no single procedure outperforming all other models, but for Ethereum, there is evidence showing that the GAS model is a suitable alternative for forecasting both risk measures. We found that the combining methods were not able to outperform the better of the individual models.

Assessing the Accuracy of Exponentially Weighted Moving Average Models for Value-at-Risk and Expected Shortfall of Crypto Portfolios

“…the VaR and ES forecasting performance of highly complex GARCH and GAS model specifications can be on par with relatively simpler models such as the standard GARCH. For instance, this is shown in the results of Bonello and Suda (2018), Troster et al (2019), Acereda et al (2020), Silahli et al (2021) and also in the working paper results of Trucíos and Taylor (2022).…”

“…More sophisticated univariate models include the realized GARCH and stochastic volatility models which are discussed by Takahashi et al (2016), Chen et al (2021) and Takahashi et al (2021). In the context of cryptocurrencies, stochastic volatility models have been used by Tiwari et al (2019) and realized GARCH by Trucíos and Taylor (2022)-but only in the univeraite context. These papers also have mixed results, suggesting a possible need for further research.…”

“…Trucíos (2019) evaluates VaR forecasts for bitcoin between 2011 and 2017 using six competing models, finding that only a robust bootstrap VaR method produces accurate forecasts at the 1% significance level. In fact, in the preliminary results of a subsequent working paper, Trucíos and Taylor (2022) use a more recent sample period and show that bitcoin and ether VaR forecasts based on simpler volatility models such as the standard GARCH may be considered accurate. Acereda et al (2020) find that more complex model specifications do not outperform the simpler ones for bitcoin VaR, as long as heavy-tailed distributions are used instead of the standard normal.…”

Forecasting value-at-risk and expected shortfall in emerging market: does forecast combination help?

“…Although forecast combination strategies have been providing several good results in several fields (see, for example, Atiya, 2020;Thomson et al, 2019), the literature of forecast combination in the VaR and ES context is surprisingly limited. Halbleib and Pohlmeier (2012), McAleer et al (2013) and Jeon and Taylor (2013) considered combining method to improve VaR forecasts using quantile regressions, whilst Taylor (2020) and Truc ıos and Taylor (2022) are few studies extended the forecast exercises to ES measure to selected developed market and cryptocurrencies. We further explored the potential improvement in forecasting ability of risk forecasting combination methods in emerging markets.…”

“…Our individual risk models include three popular specifications of the GARCH-type models, namely the original GARCH model of Bollerslev (1987) and its asymmetric volatility version GJR-GARCH model of Glosten et al (1993) and the fractionally integrated GARCH (FIGARCH) model of Baillie et al (1996), which captures the long memory of the variance process. The next forecasting method is a class of the observation-driven time series models, which is the generalized autoregressive score (GAS) model of Creal et al (2013), extended to VaR and ES forecasts by Truc ıos and Taylor (2022). We also employed three alternative semi-parametric models, including the CAViaR-regression for ES and CAViaR-EVT method proposed by Manganelli and Engle (2004) and the recently developed joint estimation of VaR and ES using asymmetric laplace density (ALD) function of Taylor (2019).…”