Skip to content

A Geometry-Aware Residual Correction of Hagan's SABR Implied Volatility Formula

Bibliographic record. Follow the original-source link for the publication.

Field Value
Primary domain Volatility
Other domains
Methods Financial Ml, Research Methods
Facets
Authors Adil Reghai, Lama Tarsissi, Gérard Biau, Alex Lipton
Published 2026-05-07
Source arXiv Quantitative Finance History
Identifiers arxiv:2605.06604
URL Open original source

Editorial synthesis

Why it matters

The approach keeps a structured SABR backbone and learns residual corrections, potentially balancing interpretability and speed for calibration systems if residual behavior is validated across regimes. (abstract:S1, abstract:S2, abstract:S3, abstract:S6, abstract:S7, abstract:S8, abstract:S9)

Main author claims

  • The authors claim a hybrid method that augments neural inputs with SABR-geometric features and trains the network to learn residual error versus Hagan’s approximation rather than raw implied volatility. (abstract:S1, abstract:S2, abstract:S5, abstract:S6)
  • They further claim improved accuracy and robustness over analytical and standard neural approaches under realistic and stressed regimes, while remaining lightweight and structurally consistent for real-time pricing and calibration. (abstract:S7, abstract:S8, abstract:S9)

Data, method, or discussion scope

Scope includes the hybrid design, residual-learning target, experimental claims in regular and stressed settings, and practical calibration suitability; specific metrics, baselines, and runtime benchmarks are absent. (abstract:S1, abstract:S2, abstract:S3, abstract:S6, abstract:S7, abstract:S8, abstract:S9)

Main limitations

No explicit error metrics or stress-scenario definitions are provided for the reported gains, limiting tolerance analysis under calibration perturbations. (abstract:S8, abstract:S9)

Relationships

  • None recorded.