Paper: arXiv 2411.19317

Authors: Bo Yuan, Damiano Brigo, Antoine Jacquier, Nicola Pede

Abstract

Deep learning methods have become a widespread toolbox for pricing and calibration of financial models. While they often provide new directions and research results, their `black box’ nature also results in a lack of interpretability. We provide a detailed interpretability analysis of these methods in the context of rough volatility - a new class of volatility models for Equity and FX markets. Our work sheds light on the neural network learned inverse map between the rough volatility model parameters, seen as mathematical model inputs and network outputs, and the resulting implied volatility across strikes and maturities, seen as mathematical model outputs and network inputs. This contributes to building a solid framework for a safer use of neural networks in this context and in quantitative finance more generally.

Complexity vs Empirical Score

  • Math Complexity: 8.0/10
  • Empirical Rigor: 3.0/10
  • Quadrant: Lab Rats — high math complexity, limited empirical testing

Why this score: The paper employs advanced mathematics including stochastic calculus for rough Heston models, complex neural network architectures, and sophisticated interpretability methods like SHAP and LIME. However, it focuses on theoretical interpretability analysis using synthetic data without reporting real-market backtests or performance metrics.

Research Flowchart

  flowchart TD
  A["Research Goal: Interpret the deep learning inverse map<br>in rough volatility model calibration"] --> B["Data Input:<br>Model Parameters & Implied Volatility Surfaces"]
  B --> C["Method: Neural Network<br>Calibration Architecture"]
  C --> D["Computational Process:<br>Forward Pass (IV Surface -> Parameters)<br>Backward Pass (Sensitivity Analysis)"]
  D --> E["Key Finding 1: Learned<br>Parameter Sensitivity Patterns"]
  D --> F["Key Finding 2: Non-linear<br>Relationships Exposed"]
  E --> G["Outcome: Enhanced Model<br>Interpretability Framework"]
  F --> G

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 detailed interpretability analysis of deep neural networks used to invert rough Heston model parameters to implied volatility surfaces.

Source sentence
We provide a detailed interpretability analysis of these methods in the context of rough volatility - a new class of volatility models for Equity and FX markets.

Assumes 2

  • The model is considered under the risk‑neutral measure and ignores interest rates and dividends. model
    Source sentence
    under the risk-neutral measure (where we ignore interest rates and dividends for simplicity)
  • The volatility driver is a fractional Brownian motion with Hurst exponent H in (0, 1/2). model
    Source sentence
    by a fractional Brownian motion with Hurst exponent in (0, 1 2).

Data

  • Synthetic implied volatility dataset synthetic unknown
    Source sentence
    To do so, we first generate a dataset of implied volatilities from several rough Heston models parameter sets

Costs

Not applicable The paper reports no trading or execution result.

Finds 4

  • The feed‑forward neural network reaches an accuracy of 99% on both training and validation sets for the scaling approach. predictive-accuracy
    Source sentence
    the obtained level of accuracy reaches 99% for both scaling approaches
  • A simple one‑hidden‑layer architecture with ELU activation achieves over 90% accuracy in calibrating the rough Heston model. predictive-accuracy
    Source sentence
    the simple one-hidden-layer architecture with ELU activation function works very well, achieving over 90% accuracy
  • Out‑of‑sample forecasts for parameters κ and ρ are unsatisfactory, with average errors 0.1279 and 0.5593 respectively. negative
    Source sentence
    The forecasts for κ and ρ are clearly unsatisfactory, with average errors 0.1279 and 0.5593
  • Local interpretability methods (LIME, DeepLIFT, LRP) and global SHAP indicate short‑maturity deep in‑the‑money volatilities provide the largest contribution to calibration. other
    Source sentence
    short‑maturity deep in‑the‑money volatilities actually provide the largest contribution overall

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.

Cited by 2 papers in the archive (2 by independent authors): Shapley-based Structural Analysis of Neural Calibration for … (uses its method or data)

Code

No repository by the authors found in the paper.

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

Cited References (73)

  1. D eep learning interpretability of the parameters to smile map in a rough volatility model
  2. Empirical analysis of rough and classical stochastic volatility models to the SPX and VIX markets (2022)
  3. Optimal whitening and decorrelation (2018)
  4. The impact of data pre-processing techniques and dimensionality reduction on the accuracy of machine learning (2019)
  5. On VIX F utures in the rough B ergomi model (2018)
  6. Rough volatility (2023)
  7. Feature scaling in support vector data description (2002)
  8. Batch normalization: Accelerating deep network training by reducing internal covariate shift (2015)
and 65 more
  1. Fractional B rownian motions, fractional noises and applications (1968)
  2. Volatility options in rough volatility models (2020)
  3. Pricing under rough volatility (2016)
  4. Simulation of fractional B rownian motion (2004)
  5. From local explanations to global understanding with explainable AI for trees (2020)
  6. Axiomatic attribution for deep networks (2017)
  7. Not just a black box: Learning important features through propagating activation differences (2016) — link
  8. Deep inside convolutional networks: Visualising image classification models and saliency maps (2013) — link
  9. Visualizing and understanding convolutional networks (2014)
  10. Towards better understanding of gradient-based attribution methods for deep neural networks (2017) — link
  11. A unified approach to interpreting model predictions (2017)
  12. On pixel-wise explanations for non-linear classifier decisions by layer-wise relevance propagation (2015)
  13. Learning important features through propagating activation differences (2017)
  14. Why should I trust you? E xplaining the predictions of any classifier (2016)
  15. Interpretable Machine Learning (2022)
  16. Keras Documentation: Adam — link
  17. A predictor-corrector approach for the numerical solution of fractional differential equations (2002)
  18. Volatility model calibration with neural networks a comparison between direct and indirect methods (2020) — link
  19. On deep calibration of (rough) stochastic volatility models (2019) — link
  20. Deep learning: I nterpretability? I mperial C ollege L ondon (2019)
  21. A closed-form solution for options with stochastic volatility with applications to bond and currency options (1993)
  22. Stock price distributions with stochastic volatility: an analytic approach (1991)
  23. Deep learning tutorial - PCA and Whitening (2014) — link
  24. Data preprocessing: Whitening or sphering in python (2020) — link
  25. Asymptotic behavior of the fractional H eston model (2018)
  26. On smile properties of volatility derivatives and exotic products: understanding the VIX skew (2022)
  27. Deep learning volatility (2019) — link
  28. Note on the inversion theorem (1951)
  29. Option valuation using the fast F ourier transform (1999)
  30. Spanning and derivative-security valuation (2000)
  31. The characteristic function of rough H eston models (2019)
  32. Perfect hedging in rough H eston models (2018)
  33. Roughening H eston (2019)
  34. Volatility is rough (2018)
  35. Interpretability in deep learning for finance: a case study for the H eston model (2021) — link
  36. Examples are not enough, learn to criticize! C riticism for interpretability (2016)
  37. Explanation in A rtificial I ntelligence: Insights from the social sciences (2019)
  38. Stochastic invariance and stochastic V olterra equations (2018)
  39. Option Valuation under Stochastic Volatility (2001)
  40. The vol smile problem (2002)
  41. Hybrid scheme for B rownian semistationary processes (2017)
  42. Multifactor approximation of rough volatility models (2019)
  43. Hierarchical adaptive sparse grids and quasi- M onte C arlo for option pricing under the rough B ergomi model (2020)
  44. Weak error rates for option pricing under linear rough volatility (2022)
  45. Functional central limit theorems for rough volatility (2024)
  46. Functional quantization of rough volatility and applications to the VIX (2023)
  47. Rough volatility, path-dependent PDEs and weak rates of convergence — → 2304.03042
  48. Refinement by reducing and reusing random numbers of the Hybrid scheme for B rownian semistationary processes (2021)
  49. On the weak convergence rate in the discretization of rough volatility models (2022)
  50. Stochastic analysis of the fractional B rownian motion (1999)
  51. Affine Volterra processes (1999)
  52. Affine rough models (2023)
  53. A comparison principle between rough and non-rough H eston models—with applications to the volatility surface (2020)
  54. A GMM approach to estimate the roughness of stochastic volatility (2023)
  55. Rough volatility (2023)
  56. A path-dependent PDE solver based on signature kernels — → 2403.11738
  57. Short-time at-the-money skew and rough fractional volatility (2017)
  58. Random neural networks for rough volatility (2023) — link
  59. Deep learning calibration of option pricing models: some pitfalls and solutions (2019) — link
  60. A neural network-based framework for financial model calibration (2019)
  61. Accuracy of deep learning in calibrating HJM forward curves (2021)
  62. Deep calibration of interest rates model (2021) — link
  63. Deep calibration with random grids — → 2306.11061
  64. Deep calibration of the quadratic rough H eston model (2022)
  65. Applying Deep Learning to Calibrate Stochastic Volatility Models — → 2309.07843

4 of 73 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.