Forecasting Value-at-Risk of cryptocurrencies using the time-varying mixture-accelerating generalized autoregressive score model

Abstract

We introduce the accelerating generalized autoregressive score (aGAS) technique into the Gaussian-Cauchy mixture model and propose a novel time-varying mixture (TVM)-aGAS model. The TVM-aGAS model is particularly suitable for processing the fat-tailed and extreme volatility characteristics of cryptocurrency returns. We then apply it to Value-at-Risk (VaR) forecasting of three cryptocurrencies, obtaining testing results that show our model possesses advantages in forecasting the density of daily cryptocurrency returns. Compared to other benchmarked models, the proposed model performs well in forecasting out-of-sample VaR. The findings underscore that our method is a useful and reliable alternative for forecasting VaR in cryptocurrencies.