Paper: arXiv 2506.22888

Authors: Jirong Zhuang, Xuan Wu

Abstract

Constructing the Implied Volatility Surface (IVS) is a challenging task in quantitative finance due to the complexity of real markets and the sparsity of market data. Structural models like Stochastic Alpha Beta Rho (SABR) model offer interpretability and theoretical consistency but lack flexibility, while purely data-driven methods such as Gaussian Process regression can struggle with sparse data. We introduce SABR-Informed Multi-Task Gaussian Process (SABR-MTGP), treating IVS construction as a multi-task learning problem. Our method uses a dense synthetic dataset from a calibrated SABR model as a source task to inform the construction based on sparse market data (the target task). The MTGP framework captures task correlation and transfers structural information adaptively, improving predictions particularly in data-scarce regions. Experiments using Heston-generated ground truth data under various market conditions show that SABR-MTGP outperforms both standard Gaussian process regression and SABR across different maturities. Furthermore, an application to real SPX market data demonstrates the method’s practical applicability and its ability to produce stable and realistic surfaces. This confirms our method balances structural guidance from SABR with the flexibility needed for market data.

Complexity vs Empirical Score

  • Math Complexity: 8.5/10
  • Empirical Rigor: 8.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs advanced mathematical frameworks including Gaussian Processes, SABR dynamics, and multi-task learning with detailed kernel formulations, which indicates high math complexity. It also demonstrates high empirical rigor through Heston-generated synthetic data tests, real SPX market data applications, and specific performance comparisons against baselines.

Research Flowchart

  flowchart TD
  A["Research Goal:<br>Improve Implied Volatility Surface<br>Construction for Sparse Data"] --> B{"Key Methodology:<br>SABR-Informed Multi-Task GP"}
  
  B --> C["Data Preparation:<br>Use Synthetic SABR Data<br>as Source Task"]
  B --> D["Data Preparation:<br>Use Sparse Market Data<br>as Target Task"]
  
  C & D --> E["Computational Process:<br>Transfer Structural Information<br>from SABR to GP via Multi-Task Learning"]
  
  E --> F["Key Findings:<br>Outperforms Standard GP & SABR<br>on Heston & Real SPX Data"]
  E --> G["Outcome:<br>Balance of Structural Guidance<br>and Market Flexibility"]

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 introduces a SABR-informed multitask Gaussian process for constructing implied volatility surfaces from sparse option quotes.

Source sentence
This study introduces a SABR-informed multitask Gaussian process for constructing implied volatility surfaces from sparse option quotes.

Assumes 5

  • Synthetic data is generated by a calibrated SABR model to capture structural patterns of volatility surfaces. model
    Source sentence
    In this paper, we used the SABR model [1] for incorporating financial theory to guide the model.
  • The Heston stochastic volatility model provides realistic ground‑truth market behavior. model
    Source sentence
    We chose the Heston stochastic volatility model [14] for this purpose, as it generates complex volatility dynamics similar to real markets.
  • Gaussian noise is added to synthetic volatilities to avoid over‑reliance on synthetic data. data
    Source sentence
    In practice, Gaussian noise ϵS ∼N(0, σ2 syn) is added to the synthetic volatilities to prevent the model from becoming overly dependent on the synthetic data.
  • The target dataset is much sparser than the source dataset. data
    Source sentence
    When the target dataset DT is much sparser than the source dataset DS
Show 1 more
  • Hierarchical regularization of task embeddings mitigates over‑fitting and negative transfer. statistical
    Source sentence
    The task embeddings eS and eT are critical. Their locations in the embedding space determine how much structural information transfers between the SABR synthetic data and market observations.

Data

  • Heston ground truth data synthetic unknown
    synthetic market option contracts
    Source sentence
    We generated synthetic “market” data using the Heston model [14]
  • Synthetic SABR dataset DS synthetic unknown
    dense synthetic option quotes
    Source sentence
    This synthetic dataset acts as a source of information to guide the learning process and prediction, especially in regions where market data is scarce.
  • Market dataset DT options unknown
    166 European call option contracts
    Source sentence
    we generated a dataset DT comprising N = 166 European call option contracts

Recognised sources: OptionMetrics IvyDB

Costs

Not applicable The paper reports no trading or execution result.

Finds 3

  • The SABR‑MTGP model achieves lower error than the single‑task GP and SABR at near‑term maturities and remains competitive at long‑term maturities while satisfying no‑arbitrage conditions. predictive-accuracy
    Source sentence
    On Heston ground truth across ten market regimes and in a case study with SPX options, the model achieves lower error than the single-task Gaussian process and SABR at near-term maturities and remains competitive at long-term maturities, while satisfying standard no-arbitrage conditions.
  • In the near‑term slice, SABR‑MTGP tracks the Heston truth closely with smooth residuals, outperforming SABR and single‑task GP. performance
    Source sentence
    The single-task GP follows the curve yet exhibits small local oscillations in the wings where observations are sparse. SABR-MTGP tracks the truth closely and its residuals are centered and smooth.
  • Butterfly spreads remain non‑negative for SABR‑MTGP, indicating consistency with static no‑arbitrage on the SPX sample. statistical
    Source sentence
    The butterfly spreads shown in Figure 11 remain nonnegative for SABR-MTGP, SABR, SSVI and cubic spline on the inspected slices, while GP and NN exhibit occasional dips near mid or long maturities.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 22 related papers). Coverage is partial, so this is neither replication nor refutation.

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 (39)

  1. Gaussian processes for machine learning (2006)
  2. Managing Smile Risk (2002)
  3. Multi-task Gaussian process prediction (2007)
  4. Kernels for vector-valued functions: A review (2012)
  5. Teaching economics to the machines (2023)
  6. A closed-form solution for options with stochastic volatility with applications to bond and currency options (1993)
  7. The pricing of options and corporate liabilities (1973)
  8. Gaussian process regression for swaption cube construction under no-arbitrage constraints (2022)
and 31 more
  1. Beyond surrogate modeling: Learning the local volatility via shape constraints — → 2212.09957
  2. Deep smoothing of the implied volatility surface (2020)
  3. Incorporating prior financial domain knowledge into neural networks for implied volatility surface prediction (2021)
  4. Operator Deep Smoothing for Implied Volatility — → 2406.11520
  5. No-Arbitrage Deep Calibration for Volatility Smile and Skewness — → 2310.16703
  6. Gaussian process regression stochastic volatility model for financial time series (2016)
  7. Machine learning for quantitative finance: fast derivative pricing, hedging and fitting (2018)
  8. Sequential design and spatial modeling for portfolio tail risk measurement (2018)
  9. Machine learning for pricing American options in high-dimensional Markovian and non-Markovian models (2020)
  10. Gaussian process imputation of multiple financial series (2020)
  11. Probabilistic machine learning for local volatility (2021)
  12. Gaussian process regression for pricing variable annuities with stochastic volatility and interest rate (2021)
  13. KrigHedge: Gaussian process surrogates for delta hedging (2021)
  14. The Effect of Regression Methods on the Performance of Options Pricing using Machine Learning (2022)
  15. A Gaussian process regression machine learning model for forecasting retail property prices with Bayesian optimizations and cross-validation (2023)
  16. On the pricing of capped volatility swaps using machine learning techniques (2024)
  17. Forecasting US inflation using Bayesian nonparametric models (2024)
  18. Investigating the price determinants of the European Emission Trading System: a non-parametric approach — → 2406.05094
  19. Gaussian process vector autoregressions and macroeconomic uncertainty (2025)
  20. Option valuation using the fast Fourier transform (1999)
  21. A simplex method for function minimization (1965)
  22. Occam’s razor (2000)
  23. Arbitrage-free implied volatility surface generation with variational autoencoders (2023)
  24. Simulation of arbitrage-free implied volatility surfaces (2023)
  25. B-spline techniques for volatility modeling (2013) — link
  26. A parsimonious arbitrage-free implied volatility parameterization with application to the valuation of volatility derivatives (2004)
  27. Arbitrage-free SVI volatility surfaces (2014)
  28. The variance gamma process and option pricing (1998)
  29. A jump-diffusion model for option pricing (2002)
  30. Pricing under rough volatility (2016)
  31. Volatility is rough (2022)

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