Paper: arXiv 2509.05911
Authors: Lijie Ding, Egang Lu, Kin Cheung
Abstract
We present a deep learning framework for pricing options based on market-implied volatility surfaces. Using end-of-day S&P 500 index options quotes from 2018-2023, we construct arbitrage-free volatility surfaces and generate training data for American puts and arithmetic Asian options using QuantLib. To address the high dimensionality of volatility surfaces, we employ a variational autoencoder (VAE) that compresses volatility surfaces across maturities and strikes into a 10-dimensional latent representation. We feed these latent variables, combined with option-specific inputs such as strike and maturity, into a multilayer perceptron to predict option prices. Our model is trained in stages: first to train the VAE for volatility surface compression and reconstruction, then options pricing mapping, and finally fine-tune the entire network end-to-end. The trained pricer achieves high accuracy across American and Asian options, with prediction errors concentrated primarily near long maturities and at-the-money strikes, where absolute bid-ask price differences are known to be large. Our method offers an efficient and scalable approach requiring only a single neural network forward pass and naturally improve with additional data. By bridging volatility surface modeling and option pricing in a unified framework, it provides a fast and flexible alternative to traditional numerical approaches for exotic options.
Complexity vs Empirical Score
- Math Complexity: 7.0/10
- Empirical Rigor: 8.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper employs advanced neural network architectures (VAE, MLP) with detailed mathematical formulation for compression and mapping, indicating high mathematical complexity. It demonstrates strong empirical rigor by using real market data from 2018-2023, constructing arbitrage-free surfaces, generating training/testing splits, and providing error analysis and performance metrics.
Research Flowchart
flowchart TD
A["Research Goal: Develop fast, scalable option pricer<br>using market-implied volatility surfaces"] --> B
subgraph B ["Data & Preprocessing"]
B1["S&P 500 Index Options<br>2018-2023"] --> B2["Construct Arbitrage-Free<br>Volatility Surfaces"]
end
B --> C
subgraph C ["Deep Learning Architecture"]
C1["VAE Encoder<br>Compresses Surface to 10-D Latent"] --> C2["Multilayer Perceptron<br>Maps Latent + Inputs to Price"]
end
C --> D["Staged Training:<br>1. VAE Training 2. MLP Mapping 3. End-to-End Fine-Tune"]
D --> E
subgraph E ["Key Outcomes"]
E1["High Accuracy<br>American & Asian Options"]
E2["Errors Near Long Maturity<br>& At-the-Money Strikes"]
E3["Efficient Forward Pass<br>Superior Scalability"]
end
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 deep learning framework that uses a variational autoencoder to compress market-implied volatility surfaces and a multilayer perceptron to price American puts and arithmetic Asian options in a single forward pass.
Source sentence
We present a deep learning framework for pricing options based on market-implied volatility surfaces.
Assumes 4
- The volatility surfaces constructed from end‑of‑day quotes are arbitrage‑free. market
Source sentence
we construct arbitrage-free volatility surfaces
- Ground‑truth option values are generated by QuantLib, assuming its numerical methods are accurate. model
Source sentence
ground truth valuations are obtained from Quantlib
- Training and testing data are independent because no volatility surfaces from the test set are used in training. data
Source sentence
The training and testing sets are independent, i.e., the neural network was not trained with any volatility surfaces from the set of testing pricing data.
- Option pricing inputs are limited to end‑of‑day quotes (no intraday dynamics). market
Source sentence
Using end-of-day S&P 500 index options quotes from 2018-2023
Data
- S&P 500 index options quotes options public
Source sentence
Using end-of-day S&P 500 index options quotes from 2018-2023, we construct arbitrage-free volatility surfaces
- Generated American put and arithmetic Asian option valuations synthetic public
Source sentence
generate training data for American puts and arithmetic Asian options using QuantLib
Costs
Not applicable
Finds 3
- The trained pricer achieves high accuracy across American and Asian options, with errors mainly near long maturities and at‑the‑money strikes. predictive-accuracy
Source sentence
The trained pricer achieves high accuracy across American and Asian options, with prediction errors concentrated primarily near long maturities and at-the-money strikes
- Neural‑network predicted prices align well with ground‑truth QuantLib prices, distributing nicely around the diagonal of perfect matches. performance
Source sentence
the neural network predicted prices align well with the ground truth prices, with all of the predicted prices distributing nicely around the diagonal line of perfect matches
- The VAE can accurately reconstruct full volatility surfaces from a 10‑dimensional latent space. theoretical
Source sentence
After we have confirmed that the VAE is able to successfully and accurately reconstruct full volatility surfaces from the latent space
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 20 related papers). Coverage is partial, so this is neither replication nor refutation.
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)
- The Pricing of Options and Corporate Liabilities (1973)
- Paul Wilmott on quantitative finance (2013)
- Stochastic calculus for finance II: Continuous-time models (2004)
- Options, Futures, and Other Derivatives (2017)
- Financial modelling with jump processes (2003)
- Monte Carlo Methods in Financial Engineering (2004)
- Dynamic asset pricing theory (2010)
- Machine learning: a probabilistic perspective (2012)
and 31 more
- Deep learning (2016)
- Machine Learning in Finance: From Theory to Practice (2020)
- Pricing Options and Computing Implied Volatilities using Neural Networks (2019)
- Supervised Deep Neural Networks (DNNs) for Pricing/Calibration of Vanilla/Exotic Options (2019)
- Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations (2019)
- Meshless methods for American option pricing through physics-informed neural networks (2023)
- Option pricing in the Heston model with physics inspired neural networks (2024)
- A deep learning based numerical PDE method for option pricing (2023)
- The application of improved physics-informed neural network (IPINN) method in finance (2022)
- Machine learning algorithms for financial asset price forecasting (2020) — link
- Neural network pricing of American put options (2020)
- Accelerated American option pricing with deep neural networks (2023)
- Deep learning for exotic option valuation (2021) — link
- Neural networks for option pricing and hedging: a literature review (2019) — link
- Option pricing with machine learning (2019)
- Machine learning in finance: the case of deep learning for option pricing (2017)
- Machine learning for quantitative finance: fast derivative pricing, hedging and fitting (2018)
- A new PDE approach for pricing arithmetic average Asian options (2001)
- Arbitrage-free SVI volatility surfaces (2014)
- Fast Derivative Valuation from Volatility Surfaces using Machine Learning (2025) — link
- Tutorial on variational autoencoders (2016) — link
- Variational autoencoder for deep learning of images, labels and captions (2016)
- Derivatives Pricing using QuantLib: An Introduction (2015)
- Computational finance using QuantLib-Python (2016)
- Variational autoencoders: A hands-off approach to volatility (2021) — link
- Arbitrage-free implied volatility surface generation with variational autoencoders (2023)
- Deep smoothing of the implied volatility surface (2020)
- Hedging in the possible presence of unspanned stochastic volatility: Evidence from swaption markets (2003)
- Probability distribution in the SABR model of stochastic volatility (2015)
- Controllable Generation of Implied Volatility Surfaces with Variational Autoencoders — → 2509.01743
- Deep learning from implied volatility surfaces (2023)
1 of 39 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.