On options-driven realized volatility forecasting: Information gains via rough volatility model

Battle Count: 69

Highly relevant to quantitative trading. The paper demonstrates that rough Heston spot volatility estimators from options data significantly improve realized volatility forecasting accuracy (9% MAE reduction, 5.6pp MDA improvement over HAR benchmark) across multiple horizons up to one month. This directly informs volatility-targeting strategies, options trading, risk management, and algorithmic execution. The directional accuracy improvement (73.6% MDA) is particularly valuable for trading strategy optimization. However, the implementation requires access to large-scale options panels and computational resources for model calibration.

Very high implementation complexity. Requires: (1) access to large-scale options panel data (3+ million contracts), (2) implementation of rough Heston model with fractional Riccati equations, (3) Markovian kernel approximation (Lifted Heston), (4) training deep neural network surrogates on millions of synthetic samples, (5) iterative two-step parametric inference with penalized NLS optimization, (6) Fourier inversion for option pricing, (7) HAR model estimation with Newey-West corrections, and (8) comprehensive out-of-sample evaluation with DM tests. The combination of non-Markovian rough volatility dynamics, deep learning calibration, and econometric forecasting makes this a multi-disciplinary implementation challenge.

The methodology is described in detail with explicit equations for the rough Heston model, parametric inference procedure, and HAR specifications. However, the deep learning surrogate architecture details are referenced to prior literature rather than fully specified. Data availability is stated as 'on request' rather than a public repository. No GitHub repository is provided. The iterative two-step procedure and hyperparameter choices (e.g., lambda_n=0.2) are documented.