Forecasting Ethereum Volatility from Stablecoin DeFi Activity
Summary
This study builds separate XGBoost models to predict next-day Ethereum volatility from activity in the USDC, USDT, and DAI lending ecosystems. Features include lagged borrowing, repayment, and flash-loan measures, plus seven-day averages. It uses time-series cross-validation and a two-stage hyperparameter search, then applies SHAP values to explain how individual features push a prediction above or below its baseline.
The reported errors are lowest for USDT, followed by USDC and DAI; the authors attribute these differences to ecosystem composition and the noisier, collateral-linked behavior of DAI. Feature-importance discussion highlights borrowing, repayments, and flash loans as useful signals at different lags. The article presents predictive associations, not evidence that DeFi activity causes volatility. Its performance claims are limited to the described data and validation setup, and the excerpt is truncated before all feature results and caveats are shown.
Key ideas
- Separate models use stablecoin-specific DeFi activity to estimate next-day Ethereum volatility.
- Lagged transaction measures and rolling averages represent short-term signals and sustained activity.
- Time-series cross-validation is used to preserve temporal order and reduce leakage risk.
- SHAP values provide local explanations for how features affect individual volatility forecasts.
- Reported errors differ by stablecoin, with USDT lowest and DAI highest in the presented results.
- Predictive associations do not establish causation or guarantee performance in other periods.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.