Paper: arXiv 2509.01743
Authors: Jing Wang, Shuaiqiang Liu, Cornelis Vuik
Abstract
This paper presents a deep generative modeling framework for controllably synthesizing implied volatility surfaces (IVSs) using a variational autoencoder (VAE). Unlike conventional data-driven models, our approach provides explicit control over meaningful shape features (e.g., volatility level, slope, curvature, term-structure) to generate IVSs with desired characteristics. In our framework, financially interpretable shape features are disentangled from residual latent factors. The target features are embedded into the VAE architecture as controllable latent variables, while the residual latent variables capture additional structure to preserve IVS shape diversity. To enable this control, IVS feature values are quantified via regression at an anchor point and incorporated into the decoder to steer generation. Numerical experiments demonstrate that the generative model enables rapid generation of realistic IVSs with desired features rather than arbitrary patterns, and achieves high accuracy across both single- and multi-feature control settings. For market validity, an optional post-generation latent-space repair algorithm adjusts only the residual latent variables to remove occasional violations of static no-arbitrage conditions without altering the specified features. Compared with black-box generators, the framework combines interpretability, controllability, and flexibility for synthetic IVS generation and scenario design.
Complexity vs Empirical Score
- Math Complexity: 8.0/10
- Empirical Rigor: 6.5/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper employs advanced deep learning architectures (VAEs, conditional VAEs) and incorporates theoretical financial constraints (no-arbitrage conditions) which require complex mathematical handling. It presents numerical experiments on synthetic data (Heston/SABR models) with quantitative metrics like regression error and arbitrage violation rates, demonstrating a strong empirical component for a methodological paper.
Research Flowchart
flowchart TD A["Research Goal: Controllable IVS Generation<br>via VAEs with Interpretable Features"] --> B["Methodology: Disentanglement & Control"] B --> C["Data: Implied Volatility Surface Dataset"] C --> D["VAE Framework: Encoder & Decoder"] D --> E["Control Mechanism: Target Features<br>Embedded via Anchor Regression"] D --> F["Residual Latent Variables<br>Preserve Shape Diversity"] E --> G["Output: Synthetic IVSs<br>with Specific Features<br>Slope, Curvature, Vol-Level"] F --> G G --> H["Outcomes: High Accuracy & Market Validity"] H --> I["Post-Generation: Latent-Space Repair<br>for No-Arbitrage Conditions"] I --> J["Key Findings: Flexible, Interpretable<br>Controllable Generation for Scenarios"]
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 controllable variational autoencoder framework that disentangles user‑specified shape features from residual latent factors to generate implied volatility surfaces with explicit, interpretable control.
Source sentence
This paper presents a deep generative modeling framework for controllably synthesizing implied volatility surfaces (IVSs) using a variational autoencoder (VAE).
Assumes 6
- The latent variables y (controllable features) and z (residual) are independent in the generative model. model
Source sentence
This independence assumption encourages a disentangled latent representation,i.e., y governs specific, interpretable shape features, while z accounts for residual variation.Notably, the dimensionality of y and z is typically much smaller than that of x.
- The residual latent variable z follows a standard Gaussian distribution. statistical
Source sentence
For tractability, the residual latent variable is assumed to follow the standard Gaussian distribution, z ∼N(0, I), (17) and the posteriors have the following Gaussian form qϕ(z | x, y) ∼N �z | µz, diag(σ2 z) , (18a) pθ(x | y, z) ∼N �µx, I .
- The regression error when extracting shape features is negligible, making y deterministic given x. model
Source sentence
Regarding the controllable variable y, we assume the regression error in Equation (10) is negligible when extracting shape features (Section 3).
- The latent dimension is fixed at 5 for all experiments. model
Source sentence
To balance these considerations, we experimented with different dimensionalities and fixed the latent dimension at 5 in all experiments.
Show 2 more
- Training data consist of 60,000 IVSs on a fixed grid of log‑moneyness and maturity points. data
Source sentence
The training dataset consists of 60,000 IVSs, each represented on a fixed grid with log-moneyness m ∈ [−0.27, 0.27] and time to maturity τ ∈[0.1, 0.6], both discretized into 28 points.
- No‑arbitrage constraints are not enforced during training; violations may arise and are handled post‑generation. model
Source sentence
In addition, any generated surface must respect no-arbitrage constraints, in particular the absence of butterfly and calendar spread arbitrage. While our controllable VAE enforces feature-level structure, arbi-trage violations can still arise when latent variables are sampled in the tails of the normal distribution.
Data
- IVS dataset (combined Heston–SABR) synthetic unknown
Source sentence
Additional details are provided in Appendix A. To assess dataset independence, we trained the model on both a Heston dataset and a combined Heston–SABR dataset. The results only for the combined dataset are presented in the paper.
- Heston IVS dataset synthetic unknown
Source sentence
Additional details are provided in Appendix A. To assess dataset independence, we trained the model on both a Heston dataset and a combined Heston–SABR dataset.
Costs
Not applicable
Finds 5
- The controllable VAE achieves precise control over three features with maximum absolute errors of ~5.5×10⁻⁴ (level), 8×10⁻⁴ (slope) and 5.4×10⁻⁴ (term‑structure). performance
Source sentence
The maximum absolute errors are approximately 5.5 × 10−4 for level, 8 × 10−4 for the slope over moneyness, and 5.4 × 10−4 for the term structure slope.
- When controlling four features, maximum absolute errors increase but remain below 1×10⁻² (e.g., 5.7×10⁻⁴ level, 4.6×10⁻³ slope, 9.4×10⁻³ curvature, 2.0×10⁻³ term‑structure). performance
Source sentence
The maximum absolute errors are approximately 5.7 × 10−4 for level, 4.6 × 10−3 for slope, 9.4 × 10−3 for curvature, and 2.0 × 10−3 for term-structure.
- No arbitrage violations are observed when controllable variables are sampled within the convex hull and z is drawn from the central Gaussian region. statistical
Source sentence
Specifically, when y lies within the convex hull of the dataset label space and z is drawn from the central region of N(0, 1), all 60,000 generated IVSs satisfy both the calendar-spread and butterfly conditions.
- When sampling controllable variables from an extended domain and z from the full Gaussian prior, about 9% of generated IVSs violate calendar‑spread or butterfly arbitrage conditions. statistical
Source sentence
When y is drawn from the extended hyper-rectangular domain while z is sampled from the full Gaussian prior, approximately 9% of the generated IVSs (5, 412 out of 60, 000) contains data points which violate either the calendar-spread condition, the butterfly condition, or both.
- The post‑generation repair algorithm corrects roughly half of the arbitrage‑violating surfaces. performance
Source sentence
For the remaining small proportion that violate no-arbitrage constraints, our post-generation repair algorithm successfully corrects approximately half.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 17 related papers). Coverage is partial, so this is neither replication nor refutation.
1 paper in the archive
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 (68)
- The moment formula for implied volatility at extreme strikes (2004)
- Dynamics of implied volatility surfaces (2002)
- The Volatility Surface: A Practitioner’s Guide (2006)
- Does net buying pressure affect the shape of implied volatility functions? (2004)
- The moment formula for implied volatility at extreme strikes (2002)
- The term structure of variance swap rates and optimal variance swap investments (2010)
- Pricing event risk: Evidence from concave implied volatility curves (2025)
- How important is the term structure in implied volatility surface modeling? Evidence from foreign exchange options (2011) — link
and 60 more
- The supply and demand of S&P 500 put options (2015)
- Deep hedging (2019)
- Denoising diffusion probabilistic models (2020)
- Generative adversarial nets (2014)
- Practical Guide to Economic Scenario Generators (2016) — link
- Computing Volatility Surfaces using Generative Adversarial Networks with Minimal Arbitrage Violations — → 2304.13128
- VolGAN: A Generative Model for Arbitrage-Free Implied Volatility Surfaces (2024) — link
- Regulation (EU) No 575/2013 of the European Parliament and of the Council on Prudential Requirements for Credit Institutions and Investment Firms (Capital Requirements Regulation) (2013)
- Closed-form implied volatility surfaces for stochastic volatility models with jumps (2021)
- What type of process underlies options? A simple robust test (2003)
- Pricing under rough volatility (2016) — link
- Venturing into uncharted territory: An extensible implied volatility surface model (2022)
- Generative Models in Finance: Market Generators, a Paradigm Shift in Financial Modeling (2025)
- Gaussian GenAI: Synthetic Market Data Generation (2025)
- Financial crises, implied volatility and stress testing (2001)
- Enhancing deep hedging of options with implied volatility surface feedback information — → 2407.21138
- The pricing of options and corporate liabilities (1973)
- A closed-form solution for options with stochastic volatility with applications to bond and currency options (1993)
- Managing Smile Risk (2002)
- The pricing of options on assets with stochastic volatilities (1987)
- Pricing with a smile (1994)
- Regime Switching and European Options (2002)
- Mathematical modeling and computation in finance: with exercises and Python and MATLAB computer codes (2019)
- On the pricing of corporate debt: The risk structure of interest rates (1974)
- Online adaptive machine learning based algorithm for implied volatility surface modeling (2019)
- Modeling the dynamics of implied volatility surfaces (2009)
- Deep local volatility (2020)
- Arbitrage-free implied volatility surface generation with variational autoencoders (2023)
- Deep weighted Monte Carlo: a hybrid option pricing framework using neural networks (2023)
- Filling in Missing FX Implied Volatilities with Uncertainties: Improving VAE -Based Volatility Imputation — → 2411.05998
- $\beta$- VAE : Learning basic visual concepts with a constrained variational framework (2017)
- Auto-encoding variational B ayes (2013) — link
- Implied stochastic volatility models (2021)
- Semi-supervised learning with deep generative models (2014)
- Disentangled variational auto-encoder for semi-supervised learning (2019)
- Capturing Label Characteristics in VAE s (2021)
- Variational autoencoder for regression: Application to brain aging analysis (2019)
- Semi-supervised deep dynamic probabilistic latent variable model for multimode process soft sensor application (2022)
- Learning disentangled representations with semi-supervised deep generative models (2017)
- Martingales Versus PDEs in Finance: An Equivalence Result with Examples (2000) — link
- Can a Random Variable Be a Deterministic Function of Other Random Variables Yet Be Independent of Them? — link
- Deep Weighted Monte Carlo: A Hybrid Option Pricing Framework Using Neural Networks (2023) — link
- Does Independence Imply Conditional Independence? — link
- Semi-supervised variational autoencoders for regression: application to soft sensors (2023)
- Variational Inference: A Review for Statisticians (2017) — link
- Disentangled Variational Auto-Encoder for Semi-supervised Learning (2019) — link
- Semi-Supervised Deep Dynamic Probabilistic Latent Variable Model for Multimode Process Soft Sensor Application (2023) — link
- Capturing Label Characteristics in VAE s (2021) — link
- Real analysis (2010)
- A Novel Pricing Method for European Options Based on Fourier-Cosine Series Expansions (2009) — link
- Jumps and Stochastic Volatility: Exchange Rate Processes Implicit in Deutsche Mark Options (1996) — link
- A Theory of the Term Structure of Interest Rates (1985)
- Variational autoencoders: A hands-off approach to volatility (2021) — link
- Deep Residual Learning for Image Recognition (2016) — link
- Deep Weighted Monte Carlo: A Hybrid Option Pricing Framework Using Neural Networks (2023) — link
- Arbitrage-Free Implied Volatility Surface Generation with Variational Autoencoders (2023) — link
- Simulation of Arbitrage-Free Implied Volatility Surfaces (2023) — link
- Malliavin calculus in finance: Theory and practice (2021)
- Arbitrage-free SVI volatility surfaces (2014)
- On the limited memory BFGS method for large scale optimization (1989)
3 of 68 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.