On Chain Activity and Seven Day Volatility Forecasts for Bitcoin and Ethereum

Abstract

This study tests whether delayed blockchain activity improves forecasts of volatility over the next seven daily returns of Bitcoin and Ethereum beyond recent price information. A fixed Coin Metrics snapshot supplies 2,922 daily observations per asset for 2018 through 2025. Ridge and random forest models are fitted on 2018 through 2021 origins, selected on 2022 through 2023 origins and evaluated on 724 origins per asset in 2024 through 2025. The target is an unannualized daily return RMS over seven future observations. In the main specification, adding activity reduces Bitcoin Ridge MAE by 2.51 percent, with a 28 day moving block bootstrap 95 percent interval of −4.22 to 7.81 percent, while the Bitcoin forest MAE rises by 1.75 percent. Ethereum MAE rises by 20.00 percent for Ridge and 14.07 percent for the forest, with both paired intervals favoring the price models. A longer activity delay retains the adverse Ethereum result. Adding activity to a volume controlled Bitcoin Ridge model reduces MAE by 9.63 percent, although that forecast remains less accurate than the original price only model. Validation selects price only Ridge for both dashboard assets. The results support context dependent information value rather than a general forecasting improvement. The accompanying dashboard presents historical forecasts and validation calibrated error bands. Because the data contain provider revisions and the outcome uses daily rather than intraday returns, the evidence is a retrospective risk monitoring evaluation with limited real time and latent volatility interpretation.