Volatility Forecasting Using Range-Based Estimators: Evidence from Five Major Cryptocurrencies

Abstract

This study evaluates the predictive efficacy of three range-based volatility estimators, Parkinson (1980), Garman and Klass (1980), and Rogers and Satchell (1991), within the Heterogeneous Auto-regressive Realized Volatility (HAR-RV) framework, applied to five principal cryptocurrency mar-kets: Bitcoin (BTC), Ethereum (ETH), Binance Coin (BNB), Ripple (XRP), and Tether USD (USDT). This study utilizes daily Open-High-Low-Close (OHLC) price data from January 2020 to December 2025, comprising approximately 2,162 observations per asset, to construct daily, weekly, and monthly realized variance components for estimating the HAR-RV model for each estimator across all five cryptocurrencies. Empirical findings indicate that lagged volatility components are statisti-cally significant across all markets, with notable variations in volatility persistence patterns by asset: the daily component is predominant for BTC and ETH, while the monthly component is most significant for XRP. Of the three estimators, Parkinson (1980) regularly attains the greatest in-sample R² values, varying from 11.3% for Ripple to 26.8% for Binance Coin, surpassing the theoret-ically superior Garman-Klass estimator. This result is ascribed to microstructure noise resulting from the lack of centralized opening auctions in continuously traded digital asset markets, which distorts open price data and diminishes the trustworthiness of estimators that utilize opening pric-es. Rogers-Satchell exhibits the worst in-sample fit across all five assets. This research presents the inaugural systematic comparison of range-based volatility estimators inside a cohesive HAR-RV framework across many prominent cryptocurrencies, offering novel data for the most appro-priate OHLC-based volatility measures for predicting cryptocurrency market volatility.