Forecasting the Volatility of Bitcoin and Ethereum

Abstract

Cryptocurrencies, particularly Bitcoin and Ethereum, have introduced new dynamics to global financial markets, most notably through their extreme price volatility. As a result, the accurate forecasting of cryptocurrency volatility has become critical for traders, investors, and regulators. This study examines the forecasting performance of two prominent time series models—the Heterogeneous Autoregressive (HAR) model and the Autoregressive Moving Average (ARMA) model—by applying them to high-frequency data from 2022. The results indicate that while the ARMA model performs reasonably well in stable market conditions, it struggles to account for the sharp volatility spikes that are common in cryptocurrency markets. In contrast, the HAR model demonstrates stronger predictive accuracy, particularly during periods of heightened volatility, as it captures the persistent and multi-scale nature of cryptocurrency price movements. These findings suggest that the HAR model is a more effective tool for forecasting volatility in highly volatile environments like those seen in the cryptocurrency market, offering valuable insights for risk management and strategic decision-making.