Forecasting Volatility of Ethereum using HAR-RV Models.

This study is focused on analyzing and forecasting the realized volatility of Ethereum using models from the Heterogeneous AutoRegressive (HAR) family, which account for the long memory of financial data. In the context of the rapidly evolving and highly volatile cryptocurrency market, where Ethereum ranks second in terms of capitalization, it is crucial to develop accurate volatility forecasting methods for risk management and trading strategy formulation. The research includes an analysis of the market from 2017 to 2023, assessing volatility through various methods including realized volatility, bipower variation, and quarticity, as well as jumps, leverage effect, and the impact of US monetary policy. The primary goal of the study is to identify the HAR model that provides the greatest forecasting power. This involves analyzing different model specifications (logarithmic and quadratic) and optimizing not only the parameters of the models themselves but also the lengths of the rolling windows in which they are considered. The study employs statistical methods such as Mean Squared Error (MSE), Quasi-Likelihood function (QLIKE), and conducts an MCS test to select the most effective models among different specifications. The scientific novelty of the research lies in the comprehensive analysis of models from this family, with more than 5.5 million models of various specifications considered, and the best achieved MAPE parameter is 28.14%. The results of the study can significantly impact the practice of trading and risk management in cryptocurrency markets, providing market participants with more accurate and reliable tools for working with volatility.