Paper: arXiv 2510.19126

Authors: Keyuan Wu, Tenghan Zhong, Yuxuan Ouyang

Abstract

We present a fast and robust calibration method for stochastic volatility models that admit Fourier-analytic transform-based pricing via characteristic functions. The design is structure-preserving: we keep the original pricing transform and (i) split the pricing formula into data-independent inte- grals and a market-dependent remainder; (ii) precompute those data-independent integrals with GPU acceleration; and (iii) approximate only the remaining, market-dependent pricing map with a small neural network. We instantiate the workflow on a rough volatility model with tempered-stable jumps tailored to power-type volatility derivatives and calibrate it to VIX options with a global-to-local search. We verify that a pure-jump rough volatility model adequately captures the VIX dynamics, consistent with prior empirical findings, and demonstrate that our calibration method achieves high accuracy and speed.

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 mathematical modeling with rough volatility, tempered-stable Lévy processes, and Fourier transforms requiring complex numerical integration, indicating high math complexity. Empirically, it includes real-world VIX option data, calibration procedures, GPU acceleration, and neural network surrogates, demonstrating strong implementation and backtest readiness.

Research Flowchart

  flowchart TD
  A["Research Goal:<br>Fast Calibration of<br>Rough Volatility Models"] --> B["Methodology: Decomposition & Precomputation"]
  B --> C{"Structure-Preserving Design"}
  C --> D["Split Pricing Formula:<br>Data-Independent Integrals"]
  C --> E["Market-Dependent Remainder"]
  D --> F["Precompute Integrals<br>with GPU Acceleration"]
  E --> G["Approximate with<br>Small Neural Network"]
  F --> H["Data Input:<br>VIX Options Market Data"]
  G --> H
  H --> I["Calibration:<br>Global-to-Local Search"]
  I --> J["Key Findings & Outcomes<br>• High Accuracy & Speed<br>• Pure-Jump Rough Model<br>Captures VIX Dynamics<br>• Efficient Framework for<br>Volatility Derivatives"]

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 fast and robust GPU‑accelerated calibration pipeline that combines precomputed data‑independent integrals with a small neural network surrogate for market‑dependent pricing, applied to a rough volatility model with tempered‑stable jumps.

Source sentence
We present a fast and robust calibration method for stochastic volatility models that admit Fourier-analytic transform-based pricing via characteristic functions. The design is structure-preserving: we keep the original pricing transform and (i) split the pricing formula into data-independent integrals and a market-dep

Assumes 5

  • The VIX index follows a pure‑jump process of infinite variation. model
    Source sentence
    Their empirical analysis reveals that the volatility process lacks a Brownian diffusion component but exhibits intense jump activity, with β approaching 2
  • Risk‑neutral pricing holds for the VIX options. model
    Source sentence
    The option pricing model is based on the assumption of risk-neutral condition, so we conduct 2 data cleaning method to remove the arbitrage prices
  • European VIX put prices must be monotonic and convex in strike (no arbitrage). data
    Source sentence
    For European VIX put options, price must be non-decreasing in the strike: Ki ≤Ki+1 ⇒ P(Ki) ≤P(Ki+1).
  • Parameter c controlling jump activity lies between 0 and 1. model
    Source sentence
    Therefore, c should be between 0 and 1.
Show 1 more
  • Synthetic training data are generated from the analytical pricing model with high‑precision numerical libraries. other
    Source sentence
    In the offline stage, we construct a high-quality training dataset by simulating option prices from an analytical pricing model implemented with high-precision numerical libraries (e.g., mpmath).

Data

  • VIX put‑option data · Bloomberg terminal options commercial
    January 2, 2025 (single trading day) · 96 VIX put options at four maturities
    Source sentence
    We use VIX option quotes from a single trading day, January 2, 2025—the first trading day of the new year following the presidential election. The dataset comprises 96 observations of VIX put options at four maturities, T ∈{20, 48, 100, 258} days.
  • Synthetic option‑price dataset synthetic unknown
    Latin‑Hypercube sampled parameter vectors (N≈5,000)
    Source sentence
    Parameter Sampling: Use Latin Hypercube Sampling to draw parameter vectors {θi}N i=1 from the joint distribution defined by the chosen bounds.

Recognised sources: Bloomberg

Costs

Not applicable The paper reports no trading or execution result.

Finds 4

  • The neural‑network surrogate achieves a validation RMSE of 3.1×10⁻⁴ while reproducing the analytical pricer. pricing-calibration
    Source sentence
    A feed-forward surrogate trained on N = 5,000 Latin-Hypercube samples accurately replicates the analytical pricer, achieving a validation RMSE of 3.1 × 10−4.
  • GPU‑accelerated pricing reduces a single price evaluation from seconds to milliseconds, enabling large synthetic datasets. performance
    Source sentence
    GPU‑accelerated pricing with CuPy. High-precision quadrature executed on the GPU reduces a single price evaluation from seconds to milliseconds, enabling the generation of large synthetic datasets.
  • The calibrated pure‑jump rough‑volatility model captures VIX dynamics, with estimated activity index β̂ = 1.715 consistent with empirical β≈1.7–1.8. statistical
    Source sentence
    Combining this with our result ˆβ = 1.715, we conclude that the VIX index on January 2, 2025, satisfies the pure-jump model assumption.
  • Calibration on the single‑day VIX data yields accurate fits but is fragile due to low‑volume, arbitrage‑violating strikes, leading to large relative errors for some OTM options. negative
    Source sentence
    many quoted strikes violated basic arbitrage bounds and exhibited extremely low trading volume, making the calibrated surface strongly fragile.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 35 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 (14)

  1. Power-type derivatives for rough volatility with jumps (2020) — link
  2. Advanced model calibration on bitcoin options (2019) — link
  3. Deep learning volatility: a deep neural network perspective on pricing and calibration in (rough) volatility models (2019) — link
  4. Volatility of volatility is (also) rough (2019) — link
  5. Hybrid scheme for Brownian semistationary processes (2015) — link
  6. Volatility is rough (2014) — link
  7. Volatility Jumps (2011) — link
  8. A Comparison of Three Methods for Selecting Values of Input Variables in the Analysis of Output From a Computer Code (2000) — link
and 6 more
  1. The Fourier-cosine method for option pricing revisited (2015)
  2. NORTHWESTERN UNIVERSITY Department of Electrical Engineering and Computer Science L BFGS B FORTRAN SUBROUTINES FOR LARGE SCALE BOUND CONSTRAINED OPTIMIZATION by (1994)
  3. Genetic Algorithms in Search, Optimization, and Learning (1989)
  4. Sample functions of stochastic processes with independent increments (1961)
  5. GPU-accelerated pricing with CuPy. High-precision quadrature executed on the GPU reduces a single price evaluation from seconds to milliseconds, enabling the generation of large synthetic datasets
  6. Parameter Sampling: Use Latin Hypercube Sampling to draw parameter vectors { θ i } Ni =1 from the joint distribution defined by the chosen bounds

0 of 14 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.