Paper: arXiv 2609.37741

Authors: Frédéric Pauquay

Abstract

We develop a non-perturbative framework for stochastic-volatility option pricing organised by the two-particle-irreducible (2PI) effective action and the Dyson-Schwinger gap equations of quantum field theory. In log-price, log-volatility or Lamperti coordinates, the joint law of the state variables is approximated by a self-consistent Gaussian whose mean and effective diffusion follow from the 2PI stationarity conditions, with the drift Jacobian given by statistical linearisation. The smile-generating exponential and CEV interactions are evaluated through the exact Gaussian moment-generating function rather than a Taylor cut. Whether the dressed inverse propagator is local in time separates Markovian models, where the gap equation collapses to a few ODEs, from rough (Volterra) models, where the full two-time propagator is retained and the characteristic function becomes a Gaussian integral over the log-variance field. Across exp-OU, SABR, rough Bergomi and rough SABR, the resulting deterministic engines match PDE or quasi-Monte-Carlo references from sub-basis-point (exp-OU) through single-digit basis points (SABR, rough Bergomi) to tens of basis points (rough SABR), with rough Heston as an exactly-transformable control. Conditional on the volatility field, forward-start smiles and continuously-monitored barriers reduce to field-only quadratures, in rough Bergomi directly on the native Volterra field, and the causal response block yields the full impulse-vega curve in one contraction at one-to-two orders of magnitude below bump-and-revalue. A technical supplement with complete derivations, extended benchmarks and secondary applications (stochastic-rate FX local volatility, quadratic Gaussian Volterra variance and the arbitrage-free FX triangle) is provided as an ancillary file.

Complexity vs Empirical Score

  • Math Complexity: 9.5/10
  • Empirical Rigor: 7.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: This paper presents an exceptionally complex and novel mathematical framework derived from quantum field theory for option pricing. While the empirical validation is strong with benchmarks, the primary contribution lies in the theoretical development and its application to rough volatility models. The clarity is good given the complexity, but the specialized jargon makes it challenging for a broader quant finance audience.

Research Flowchart

  flowchart TD
    A["Research Goal: Non-perturbative Stochastic Volatility Option Pricing"] --> B{"Methodology: 2PI Effective Action & Dyson-Schwinger Equations"};
    B --> C{"Key Concepts: Self-consistent Gaussian Approx., Statistical Linearization, Exact Gaussian Moment-Generating Function"};
    C --> D["Model Types & Inputs: Exp-OU, SABR, Rough Bergomi, Rough SABR, Rough Heston"];
    D --> E{"Computational Process: Solving Gap Equations (ODEs/Full Propagator) & Characteristic Function via Gaussian Integral"};
    E --> F["Outcomes: High Accuracy Engine (sub-bps to tens of bps), Efficient Exotics Pricing (Forward-Start Smiles, Barriers), Fast Impulse-Vega"];

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 non‑perturbative framework for stochastic‑volatility option pricing based on the two‑particle‑irreducible (2PI) effective action and Dyson‑Schwinger gap equations.

Source sentence
We develop a non-perturbative framework for stochastic-volatility option pricing organised by the two-particle-irreducible (2PI) effective action and the Dyson-Schwinger gap equations of quantum field theory.

Assumes 5

  • The joint law of state variables can be approximated by a self‑consistent Gaussian whose mean and effective diffusion follow from 2PI stationarity conditions. model
    Source sentence
    In log-price, log-volatility or Lamperti coordinates, the joint law of the state variables is approximated by a self-consistent Gaussian whose mean and effective diffusion follow from the 2PI stationarity conditions
  • The dressed inverse propagator is either local in time (Markovian) or non‑local (rough) which determines the form of the gap equations. model
    Source sentence
    Whether the dressed inverse propagator is local in time separates Markovian models, where the gap equation collapses to a few ODEs, from rough (Volterra) models, where the full two-time propagator is retained
  • Statistical linearisation is used for the drift Jacobian, replacing the gradient at the mean by its Gaussian average. statistical
    Source sentence
    The response is R = (∂t −J)−1, in which the drift gradient at the mean, ∇µ( ¯ϕ), is replaced by its Gaussian average J = ESCG[∇µ]; this is statistical linearisation
  • The discounted forward price is a martingale, imposing φ(−i)=1 as a no‑arbitrage condition. model
    Source sentence
    One exact constraint must survive any truncation: the discounted forward is a martingale, so an admissible closure preserves φ(−i) = 1, imposed as a no-arbitrage condition on the mean equation
Show 1 more
  • Synthetic benchmarks (PDE, quasi‑Monte‑Carlo, Sobol Monte Carlo) are used as ground truth for validation. data
    Source sentence
    Against a tower PDE and a 220-path Sobol reference the SCG achieves 0.1–0.2 bp RMSE across all strikes and maturities

Data

  • Synthetic PDE and quasi‑Monte‑Carlo references for exp‑OU, SABR, rough Bergomi, rough SABR synthetic public
    Source sentence
    Across the main non-affine test cases—exp-OU, SABR, rough Bergomi and rough SABR—the resulting deterministic engines match PDE or quasi-Monte-Carlo references
  • Sobol Monte Carlo benchmark (220‑path reference) benchmark public
    Source sentence
    Against a tower PDE and a 220-path Sobol reference the SCG achieves 0.1–0.2 bp RMSE across all strikes and maturities
  • PCA‑ordered Sobol field ensemble for rough Bergomi benchmark public
    Source sentence
    The outer integral over the field is done by PCA-ordered Sobol; because the engine runs on the native two-time covariance, one field ensemble is amortised across strikes

Costs

Not applicable The paper reports no trading or execution result.

Finds 5

  • The deterministic engines achieve sub‑basis‑point accuracy for exp‑OU (0.1–0.2 bp), single‑digit basis points for SABR and rough Bergomi, and tens of basis points for rough SABR. pricing-calibration
    Source sentence
    the resulting deterministic engines match PDE or quasi-Monte-Carlo references from sub-basis-point (exp-OU) through single-digit basis points (SABR, rough Bergomi) to tens of basis points (rough SABR)
  • Rough Heston control reproduces option prices within 1–5 bp, validating the non‑local propagator machinery. pricing-calibration
    Source sentence
    reproduces option prices/implied volatilities obtained from the transform to within 1–5 bp, validating the non-local propagator machinery
  • Rough Bergomi PCA‑resolved conditional RQMC achieves RMSE of ~1–3 bp across coupling regimes, while the Level‑1 Gaussian closure incurs errors of tens to hundreds of basis points. pricing-calibration
    Source sentence
    Table 3 places the accuracy against Sobol Monte Carlo: the PCA-resolved conditional RQMC holds to ∼1–3 bp across mild to extreme coupling, while the diagonal Gaussian/SCG closure
  • The causal response block yields the full latent‑factor impulse‑vega curve in one contraction at one‑to‑two orders of magnitude below bump‑and‑revalue. other
    Source sentence
    the causal response block yields the full latent-factor impulse-vega curve in one contraction at one-to-two orders of magnitude below
  • The framework provides a single engine that prices forward‑start smiles and continuously‑monitored barriers directly on the native two‑time covariance without spot‑path simulation or Markovian lift. other
    Source sentence
    a single engine that prices finite-maturity forward-start smiles and continuously-monitored touches and knock-outs directly on the native two-time covariance of a Gaussian–Volterra field, with no spot-path simulation and no Markovian lift

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.

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.