Power of decomposition in volatility forecasting for Bitcoins

This article aims to show the power of decomposition techniques in estimating Bitcoin returns volatility with a time series volatility model. The realized generalized autoregressive heteroscedasticity (RGARCH) model, employing high-frequency data in the form of realized measures, is integrated with empirical mode decomposition (EMD) and variational mode decomposition (VMD) to estimate volatility. The high fluctuations in Bitcoin prices suggest using jump-robust estimators. The superior forecasting accuracy of proposed models compared to RGARCH and GARCH models across various metrics underscores the utility of decomposition in the volatility modeling of Bitcoin returns. VMD reigns supreme over EMD as it preserves the estimators’ ranking. In particular, the RGARCH-VMD model estimated using jump-robust estimators, namely realized tri-power variation and realized bi-power variation, outperforms all competing models. Since the Chicago Mercantile Exchange officially offers Bitcoin options, the strong performance of our models can be valuable for option pricing and risk management.