Paper: arXiv 2406.11520

Authors: Ruben Wiedemann, Antoine Jacquier, Lukas Gonon

Abstract

We devise a novel method for nowcasting implied volatility based on neural operators. Better known as implied volatility smoothing in the financial industry, nowcasting of implied volatility means constructing a smooth surface that is consistent with the prices presently observed on a given option market. Option price data arises highly dynamically in ever-changing spatial configurations, which poses a major limitation to foundational machine learning approaches using classical neural networks. While large models in language and image processing deliver breakthrough results on vast corpora of raw data, in financial engineering the generalization from big historical datasets has been hindered by the need for considerable data pre-processing. In particular, implied volatility smoothing has remained an instance-by-instance, hands-on process both for neural network-based and traditional parametric strategies. Our general operator deep smoothing approach, instead, directly maps observed data to smoothed surfaces. We adapt the graph neural operator architecture to do so with high accuracy on ten years of raw intraday S&P 500 options data, using a single model instance. The trained operator adheres to critical no-arbitrage constraints and is robust with respect to subsampling of inputs (occurring in practice in the context of outlier removal). We provide extensive historical benchmarks and showcase the generalization capability of our approach in a comparison with classical neural networks and SVI, an industry standard parametrization for implied volatility. The operator deep smoothing approach thus opens up the use of neural networks on large historical datasets in financial engineering.

Complexity vs Empirical Score

  • Math Complexity: 8.5/10
  • Empirical Rigor: 7.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs advanced mathematics including neural operators, graph neural operators, and PDE-related theory, reflecting high complexity. It demonstrates strong empirical rigor by backtesting on 10 years of raw intraday S&P 500 options data, comparing against industry standards (SVI) and showcasing robustness, though it lacks explicit code or live trading implementation details.

Research Flowchart

  flowchart TD
  A["Research Goal<br>Implied Volatility Smoothing"] --> B["Data Input<br>Raw S&P 500 Options Data"]
  B --> C["Methodology<br>Graph Neural Operator"]
  C --> D["Computational Process<br>Operator Mapping"]
  D --> E["Key Outcome 1<br>Single Model Instance"]
  D --> F["Key Outcome 2<br>Adheres to No-Arbitrage Constraints"]
  D --> G["Key Outcome 3<br>Robust to Subsampling"]
  E --> H["Final Result<br>Operator Deep Smoothing"]
  F --> H
  G --> H

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper proposes Operator Deep Smoothing, a neural‑operator based method that directly maps observed option data to smoothed implied volatility surfaces.

Source sentence
We introduce operator deep smoothing, a general approach for discretization‑invariant data interpolation based on neural operators, and apply it to implied volatility smoothing.

Assumes 5

  • The implied volatility surface satisfies the strike arbitrage constraint (strictly positive implied densities). model
    Source sentence
    The inclusion of ε promotes strictly positive implied densities (we will use ε = 10⁻³)
  • The calendar arbitrage constraint holds on a bounded domain D away from zero time‑to‑expiry. model
    Source sentence
    ... where again we ignore the asymptotic condition since D is bounded away from zero time‑to‑expiry.
  • Option data are observed on a bounded domain D = (0.01,1)×(−1.5,0.5) in transformed coordinates. data
    Source sentence
    setting D = (ρmin, ρmax) × (zmin, zmax) = (0.01, 1) × (−1.5, 0.5).
  • The raw intraday S&P 500 options data are sufficiently rich to train a single model instance for the whole ten‑year period. market
    Source sentence
    We adapt the graph neural operator architecture to do so with high accuracy on ten years of raw intraday S&P 500 options data, using a single model instance.
Show 1 more
  • Training loss based on root‑mean‑square relative error yields smooth, scale‑invariant outputs. statistical
    Source sentence
    we suggest a fitting loss based on the root mean square relative error, Lfit(θ, v)

Data

Backtest sample: US equity options · options · intraday

  • CBOE S&P 500 Index Option data (20‑minute intervals) · CBOE options commercial
    2012‑2021 · S&P 500 index options
    Source sentence
    We perform our numerical experiments using 20‑minute interval CBOE S&P 500 Index Option data from 2012 to 2021. The dataset amounts to a collection of 49089 implied volatility surfaces and just above 60 million individual volatility datapoints.

Recognised sources: Cboe WRDS

Costs

Not applicable The paper reports no trading or execution result.

Finds 3

  • The operator deep smoothing method achieves lower absolute and spread‑adjusted errors than the industry‑standard SVI benchmark on both training and out‑of‑sample test periods. performance
    Source sentence
    Figure 3: Benchmark of our method (OpDS) vs. SVI, split between training period (left half; 2012 to 2020) and testing period (right half; 2021). Top and middle: Surface‑averages ⟨δabs⟩ and ⟨δspr⟩; shaded regions depict respective interquartile range.
  • The trained operator respects no‑arbitrage constraints and remains robust to subsampling of input quotes. theoretical
    Source sentence
    The trained operator adheres to critical no‑arbitrage constraints and is robust with respect to subsampling of inputs (occurring in practice in the context of outlier removal).
  • The approach generalizes to end‑of‑day data for three additional US indices without any training on those indices. performance
    Source sentence
    We proceed to successfully demonstrate the generalization capabilities of our model for end‑of‑day options data of the S&P 500 as well as three further major US indices. No data from these three indices has been used for training.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 14 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 7 papers in the archive (7 by independent authors)

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.

Cited References (39)

  1. Deep smoothing of the implied volatility surface (2020)
  2. Deep calibration with random grids — → 2306.11061
  3. Variational autoencoders: A hands-off approach to volatility (2021) — link
  4. Cboe GLOBAL MARKETS REPORTS TRADING VOLUME FOR DECEMBER AND FULL YEAR 2023 (2024)
  5. Deep local volatility (2020)
  6. Detecting and Repairing Arbitrage in Traded Option Prices (2020) — link
  7. Robust calibration and arbitrage-free interpolation of SSVI slices (2019)
  8. The normalizing transformation of the implied volatility smile (2012)
and 31 more
  1. Arbitrage-free SVI volatility surfaces (2014)
  2. A parsimonious arbitrage-free implied volatility parameterization with application to the valuation of volatility derivatives (2004)
  3. Generalized arbitrage-free SVI volatility surfaces (2016)
  4. GNOT : A general neural operator transformer for operator learning (2023)
  5. Imperial College Research Computing Service ( http://doi.org/10.14469/hpc/2232 ) (2017) — link
  6. The extended SSVI volatility surface (2017)
  7. Model calibration with neural networks (2016)
  8. Deep learning volatility: a deep neural network perspective on pricing and calibration in (rough) volatility models (2021)
  9. An operator learning perspective on parameter-to-observable maps (2024) — link
  10. Deep Learning Calibration of Option Pricing Models: Some Pitfalls and Solutions (2019) — link
  11. Let’s be rational (2015)
  12. Neural operator: Learning maps between function spaces with applications to PDEs (2023)
  13. The nonlocal neural operator: Universal approximation (2023) — link
  14. Fourier neural operator for parametric partial differential equations (2020) — link
  15. Neural operator: Graph kernel network for partial differential equations (2020)
  16. A structured matrix method for nonequispaced neural operators (2023) — link
  17. Physics-informed neural operator for learning partial differential equations (2021)
  18. A neural network-based framework for financial model calibration (2019)
  19. Normalizing Volatility Transforms and Parameterization of Volatility Smile (2021)
  20. DeepONet : Learning Nonlinear Operators for Identifying Differential Equations Based on the Universal Approximation Theorem of Operators (2021) — link
  21. Options Are the Hottest Trade on Wall Street (2023)
  22. Arbitrage free implied volatility surfaces (2010)
  23. Factorized F ourier neural operators (2021) — link
  24. Machine learning for option pricing: an empirical investigation of network architectures — → 2307.07657
  25. Machine Learning and Option Implied Information (2018)
  26. No arbitrage SVI (2022)
  27. No arbitrage global parametrization for the eSSVI volatility surface (2022)
  28. Refined analysis of the no-butterfly-arbitrage domain for SSVI slices (2023)
  29. PyTorch: An Imperative Style, High-Performance Deep Learning Library (2019) — link
  30. Simulation of Arbitrage-Free Implied Volatility Surfaces (2023) — link
  31. NN De- Americanization : A Fast and Efficient Calibration Method for American-Style Options (2023) — link

2 of 39 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.