Paper: arXiv 2511.07571

Authors: Chen Jin, Ankush Agarwal

Abstract

We introduce a conditional Denoising Diffusion Probabilistic Model (DDPM) for generating arbitrage-free implied volatility (IV) surfaces, offering a more stable and accurate alternative to existing GAN-based approaches. To capture the path-dependent nature of volatility dynamics, our model is conditioned on a rich set of market variables, including exponential weighted moving averages (EWMAs) of historical surfaces, returns and squared returns of underlying asset, and scalar risk indicators like VIX. Empirical results demonstrate our model significantly outperforms leading GAN-based models in capturing the stylized facts of IV dynamics. A key challenge is that historical data often contains small arbitrage opportunities in the earlier dataset for training, which conflicts with the goal of generating arbitrage-free surfaces. We address this by incorporating a standard arbitrage penalty into the loss function, but apply it using a novel, parameter-free weighting scheme based on the signal-to-noise ratio (SNR) that dynamically adjusts the penalty’s strength across the diffusion process. We also show a formal analysis of this trade-off and provide a proof of convergence showing that the penalty introduces a small, controllable bias that steers the model toward the manifold of arbitrage-free surfaces while ensuring the generated distribution remains close to the real-world data.

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 features advanced mathematics, including diffusion processes, SDEs, and a formal convergence proof, while also demonstrating strong empirical rigor with a comparative benchmark (GAN), backtest-ready code, and quantitative metrics like MAPE and confidence intervals.

Research Flowchart

  flowchart TD
  A["Research Goal: <br>Forecast IV Surface<br>Arbitrage-Free with DDPM"] --> B{"Data & Inputs"}
  B --> B1["Market Variables<br>EWMA of IV, Returns,<br>Squared Returns, VIX"]
  B --> B2["Historical IV Surfaces<br>Contains Minor Arbitrage"]
  
  B1 & B2 --> C["Conditional DDPM Framework"]
  
  C --> D{"Arbitrage Penalty<br>in Loss Function"}
  D --> D1["Parameter-Free Weighting<br>Based on Signal-to-Noise Ratio"]
  
  C --> E["Computational Training<br>Stochastic Gradient Descent"]
  E --> F
  
  subgraph F ["Key Methodological Innovation"]
      D1 --> F1["Dynamic Steering<br>Toward Arbitrage-Free Manifold"]
  end
  
  F1 --> G["Outcomes & Findings"]
  G --> G1["Significantly Outperforms<br>GAN-based Models"]
  G --> G2["Stylized Facts<br>of IV Dynamics Captured"]
  G --> G3["Formal Proof:<br>Bias-Variance Trade-off"]

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 a conditional Denoising Diffusion Probabilistic Model (DDPM) to generate one‑day‑ahead arbitrage‑free implied volatility surfaces.

Source sentence
We introduce a conditional Denoising Diffusion Probabilistic Model (DDPM) for generating one-day ahead arbitrage-free implied volatility surfaces.

Assumes 3

  • The target surface for the next trading day is drawn from a distribution conditioned on the current market state Y(k). model
    Source sentence
    we assume that the target surface for the upcoming trading day k + 1 is drawn from a distribution governed entirely by the current conditioning set, denoted as pdata(· | Y(k))
  • Volatility dynamics follow a Variance Preserving (VP) stochastic differential equation. model
    Source sentence
    selecting a Variance Preserving (VP) SDE appropriate for the volatility data
  • Market conditioning variables include exponentially weighted moving averages of historical vol‑surfaces, returns, squared returns and scalar risk indicators. data
    Source sentence
    we condition our model on a set of market variables, including exponentially weighted moving averages (EWMAs) of historical vol-surfaces, returns and squared returns of the underlying asset, and scalar risk indicators associated with the underlying asset

Data

  • OptionMetrics option price data · OptionMetrics options commercial
    1996-01-04 to 2023-08-31
    Source sentence
    real market data from the Option Prices file provided by OptionMetrics (OptionMetrics, 2023), spanning the period from January 4, 1996, to August 31, 2023

Recognised sources: OptionMetrics IvyDB

Costs

Not applicable The paper reports no trading or execution result.

Finds 3

  • The diffusion model achieves an overall MAPE of 3.0026%, lower than the benchmark's 3.8418%. performance
    Source sentence
    Our diffusion model demonstrates strong predictive accuracy, achieving an overall MAPE of 3.0026%, which is lower than the 3.8418% from the benchmark
  • The diffusion model attains MAE of 0.7062% and RMSE of 1.0814%, outperforming the benchmark MAE of 0.9120% and RMSE of 1.3027%. predictive-accuracy
    Source sentence
    Our model achieves an overall Mean Absolute Error (MAE) of 0.7062% and a Root Mean Square Error (RMSE) of 1.0814%, outperforming the benchmark’s MAE of 0.9120% and RMSE of 1.3027%
  • The model is better calibrated than the benchmark generative model. pricing-calibration
    Source sentence
    Our model not only achieved a lower overall MAPE than the benchmark generative model but also was shown to be better calibrated

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 24 related papers). Coverage is partial, so this is neither replication nor refutation.

Code

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

Cited References (58)

  1. Score-Based Generative Modeling through Stochastic Differential Equations (2021) — link
  2. Stochastic Volatility Modeling (2016)
  3. Prices of State-Contingent Claims Implicit in Option Prices (1978)
  4. Arbitrage-free SVI volatility surfaces (2014)
  5. Incorporating prior financial domain knowledge into neural networks for implied volatility surface prediction (2021)
  6. Optuna: A Next-generation Hyperparameter Optimization Framework (2019)
  7. Forecasting Implied Volatility Smile Surface via Deep Learning and Attention Mechanism (2019)
  8. Diffusion Models for Adaptive Sequential Data Generation (2026) — link
and 50 more
  1. Operator deep smoothing for implied volatility — → 2406.11520
  2. DYSANOS Generative Dynamic Smooth Arbitrage-free Non-parametric Option Surfaces (2026) — link
  3. Arbitrage-free implied volatility surface generation with variational autoencoders (2023)
  4. FuNVol : multi-asset implied volatility market simulator using functional principal components and neural SDEs (2024)
  5. Long Short-Term Memory (1997) — link
  6. Managing Smile Risk (2002)
  7. Reverse-time diffusion equation models (1982)
  8. A Connection Between Score Matching and Denoising Autoencoders (2011)
  9. Deep Smoothing of the Implied Volatility Surface (2020)
  10. Physics-Informed Diffusion Models (2025)
  11. Constrained Synthesis with Projected Diffusion Models (2024)
  12. From Implied to Spot Volatilities (2010)
  13. Implied Volatility Functions: Empirical Tests (1998)
  14. The Dynamics of the S&P 500 Implied Volatility Surface (2000)
  15. A Semiparametric Factor Model for Implied Volatility Surface Dynamics (2007)
  16. Predictable Dynamics in the S&P 500 Index Options Implied Volatility Surface (2006)
  17. Can We Forecast the Implied Volatility Surface Dynamics of Equity Options? P redictability and Economic Value Tests (2014)
  18. Evaluating Interval Forecasts (1998)
  19. Testing the Equality of Prediction Mean Squared Errors (1997)
  20. A Simple, Positive Semi-Definite, Heteroskedasticity and Autocorrelation Consistent Covariance Matrix (1987)
  21. Denoising Diffusion Implicit Models (2021)
  22. Arbitrage free implied volatility surfaces (2010)
  23. Sigmoid-weighted linear units for neural network function approximation in reinforcement learning (2018)
  24. Conditional score-based diffusion models for Bayesian inference in infinite dimensions (2023)
  25. Efficient Diffusion Training via Min- SNR Weighting Strategy (2023)
  26. Conditional Diffusion Models are Minimax-Optimal and Manifold-Adaptive for Conditional Distribution Estimation (2024)
  27. Comparing Predictive Accuracy (1995)
  28. Arbitrage-free SVI volatility surfaces (2014) — link
  29. VolGAN : A Generative Model for Arbitrage-Free Implied Volatility Surfaces (2024)
  30. Dynamics of implied volatility surfaces (2002)
  31. FiLM : Visual Reasoning with a General Conditioning Layer (2018)
  32. U-Net : Convolutional Networks for Biomedical Image Segmentation (2015)
  33. Attention is All you Need (2017)
  34. Pricing and calibration in the 4-factor path-dependent volatility model — → 2406.02319
  35. Decoupled Weight Decay Regularization (2019)
  36. Variational Autoencoders: A Hands-Off Approach to Volatility (2022)
  37. Denoising Diffusion Probabilistic Models (2020)
  38. Diffusion Factor Models: Generating High-Dimensional Returns with Factor Structure — → 2504.06566
  39. Improved Denoising Diffusion Probabilistic Models (2021)
  40. DiffWave : A Versatile Diffusion Model for Audio Synthesis (2021)
  41. Diffusion Models Beat GANs on Image Synthesis (2021)
  42. Autoregressive Denoising Diffusion Models for Multivariate Probabilistic Time Series Forecasting (2021)
  43. Deep Unsupervised Learning using Nonequilibrium Thermodynamics (2015)
  44. Generative Modeling by Estimating Gradients of the Data Distribution (2019)
  45. A Closed-Form Solution for Options with Stochastic Volatility with Applications to Bond and Currency Options (1993)
  46. Managing Smile Risk (2002)
  47. Pricing with a smile (1994)
  48. Option valuation using the fast Fourier transform (1999)
  49. The Pricing of Options and Corporate Liabilities (1973)
  50. [dataset] Ivy DB US (2023) — link

3 of 58 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.