Rough Volatility Across Assets

Battle Count: 92

Highly relevant for quantitative trading. The paper provides cross-asset evidence that volatility is rough (H far below 0.5) across all major asset classes, which directly impacts volatility forecasting, options pricing, risk management, and trading strategy design. The failure taxonomy identifies where rough-volatility models apply and where they break down (rates, FX, seasonal commodities), which is critical for model selection in trading systems. The implied vs. realized comparison informs how to calibrate models from market data. The heterogeneity across commodities suggests asset-specific model calibration is necessary.

Moderate to high complexity. Requires: (1) high-frequency intraday data processing (1-minute or 1-second), (2) implementation of seven different realized volatility estimators with specific tuning parameters, (3) OLS regression on log-log grids with lag-window selection, (4) two-estimator offset correction with joint optimization over (H, nu, omega), (5) ATM skew computation from option chains with put-call parity regressions, (6) Black-76 and BAW implied volatility calculations, (7) Monte Carlo simulation of fBm/fOU paths via circulant embedding. The statistical framework is well-defined but the data infrastructure (multiple feeds, roll rules, quality screens) adds engineering complexity.

The paper uses commercial data (Databento, Nasdaq ITCH, CME GLBX MDP 3.0, OPRA) which is not freely available. The methodology is well-documented with explicit formulas for all seven realized volatility estimators, the two-estimator correction, and the ATM skew regression. Simulation design is fully specified (1000 paths, 10 years, specific noise levels). However, no code repository is mentioned, and the proprietary data pipeline limits full reproducibility.