Paper: arXiv 2611.19792
Authors: Saurabh Bansal, Srikanth K. Iyer, Shashi Jain
Abstract
We study a random-feature method for pricing and calibrating European options under jump-diffusion (Merton, Kou) and stochastic-volatility (Heston, SABR) models. The network has one hidden layer with a probit activation and fixed random inner weights, so fitting it to the pricing equation, the payoff and the boundary conditions is a single regularised least-squares problem, with no iterative training. For Merton and Kou the jump integral of each feature is explicit, so no quadrature is needed. One factorisation of the least-squares matrix gives the price at every state and time, its Greeks, and the exact derivatives of the prices with respect to the model parameters for gradient-based calibration. For the jump models, subtracting the Black-Scholes price removes the payoff kink and reduces the pricing error by about two orders of magnitude. The method is then about as accurate as a standard Crank-Nicolson finite-difference scheme at the same run time, although a finite-difference scheme that also subtracts the Black-Scholes price is about two orders of magnitude more accurate still. On the test problems of a recent physics-informed neural network study, the method matches that network’s accuracy in seconds instead of about half an hour. The calibration experiments show that a solver can fit the quotes and still return wrong parameters when its pricing error is not small compared with the effect of the parameters on the prices: without the subtraction the solver misses the jump parameters by up to 50%, and with it they are recovered to within 3%. For Heston and SABR, used as an applicability study with boundary conditions that need no reference prices, pricing errors are about one percent and the calibrated parameters are within about 15% of the true values.
Complexity vs Empirical Score
- Math Complexity: 8.0/10
- Empirical Rigor: 7.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper presents a mathematically sophisticated random-feature method for option pricing and calibration, demonstrating strong empirical validation against established benchmarks. Its novel approach to handling jump terms and efficient calibration process contributes significantly to the field.
Research Flowchart
flowchart TD
A["Research Goal: Price & Calibrate European Options under Jump-Diffusion & SV Models"] --> B{"Key Methodology: Random-Feature Method"};
B --> C["Method Details: 1 Hidden Layer, Probit Activation, Fixed Random Inner Weights"];
C --> D["Problem Formulation: Single Regularised Least-Squares Problem"];
D --> E["Computational Processes: Explicit Jump Integral for Merton/Kou, Matrix Factorization for Prices/Greeks/Parameter Derivatives"];
E --> F{"Key Finding 1: Subtracting Black-Scholes Price Reduces Error by 2 Orders of Magnitude for Jump Models"};
F --> G["Key Finding 2: Comparable Accuracy to Crank-Nicolson, Matches PINN accuracy in Seconds"];
G --> H["Key Finding 3: Improved Calibration Accuracy (Jump Params within 3% with subtraction vs. 50% without); Heston/SABR pricing errors ~1%, calibrated params within 15%"]
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 random-feature method for pricing and calibrating European options under jump‑diffusion (Merton, Kou) and stochastic‑volatility (Heston, SABR) models.
Source sentence
We study a random-feature method for pricing and calibrating European options under jump-diffusion (Merton, Kou) and stochastic-volatility (Heston, SABR) models.
Assumes 5
- The interest rate r is constant. model
Source sentence
We consider a European derivative on this asset that pays g(ST ) at maturity T, for a given payoff function g, and denote by V (S, t) its price at time t when St = S. For ease of notation, we take the interest rate r to be constant, so the bank account is Bt = ert.
- Pricing is performed under the risk‑neutral measure Q. model
Source sentence
For the underlying, this means that e−(r−q)tSt is a martingale, and the price of the derivative is the expected discounted payoff, V (S, t) = EQ e−r(T −t)g(ST ) St = S .
- The jump integral of each random feature can be evaluated exactly for Merton and Kou models. model
Source sentence
The jump term is exact. Averaging a probit feature over a normal or a double-exponential jump gives an explicit formula (Lemma 4.3).
- A Black‑Scholes price subtraction is applied to remove the payoff kink. other
Source sentence
3. For the jump models, a quadrature-free Black–Scholes subtraction removes the main source of error.
Show 1 more
- The underlying asset may pay a continuous dividend yield q ≥ 0. model
Source sentence
Let St be the price at time t of the underlying asset, which pays a continuous dividend yield q ≥0 (for a currency, q is the foreign interest rate).
Data
- Synthetic test problems for Merton, Kou, Heston and SABR pricing and calibration synthetic unknown
Source sentence
Experiment Model parameters Domain m nens Merton pricing r = 0.05, T = 1, σ = 0.2, λ = 0.1, µJ = −0.1, σJ = 0.3 x ∈[−1.2, 1.2], τ ∈[0, 1]; finite differences on [−2.5, 2.5] 1800 4 Kou pricing r = 0.05, T = 1, σ = 0.2, λ = 0.3, p = 0.4, η1 = 10, η2 = 5 x ∈[−1.2, 1.2], τ ∈[0, 1]; finite differences on [−2.5, 2.5] 1800 4 Jump calibration as for pricing x ∈[−1.2, 1.2], τ ∈[0, 1] 800 1 (5 in the repeat study) PINN
- Monte Carlo simulation reference for SABR synthetic unknown
Source sentence
On a separate grid with maturities up to five years and strikes up to K/F0 = 2, compared with Monte Carlo simulation of (3.8) (400,000 paths, 500 time steps), PIRanNN is at least as close to the Monte Carlo price as Hagan’s formula at 9 of the 10 points. We simulate α exactly on a time grid, with 8 million paths, and the standard error is below 0.01% of the price at every quote.
Costs
Not applicable
Finds 5
- The method matches the accuracy of a recent physics‑informed neural network in seconds instead of about half an hour. performance
Source sentence
On the test problems of a recent physics-informed neural network study, the method matches that network's accuracy in seconds instead of about half an hour.
- Subtracting the Black‑Scholes price reduces the pricing error for jump models by roughly two orders of magnitude. pricing-calibration
Source sentence
For the jump models, subtracting the Black-Scholes price removes the payoff kink and reduces the pricing error by about two orders of magnitude.
- Without the subtraction the calibration misses jump parameters by up to 50 %; with it the parameters are recovered within 3 %. pricing-calibration
Source sentence
The calibration experiments show that a solver can fit the quotes and still return wrong parameters when its pricing error is not small compared with the effect of the parameters on the prices: without the subtraction the solver misses the jump parameters by up to 50%, and with it they are recovered to within 3%.
- For Heston and SABR the pricing errors are about one percent and calibrated parameters are within about 15 % of the true values. pricing-calibration
Source sentence
For Heston and SABR, used as an applicability study with boundary conditions that need no reference prices, pricing errors are about one percent and the calibrated parameters are within about 15% of the true values.
- PIRanNN with subtraction attains an RMSE of 4.5 × 10⁻⁷ for Merton and 1.9 × 10⁻⁶ for Kou, comparable to a Crank‑Nicolson finite‑difference scheme. pricing-calibration
Source sentence
PIRanNN with subtraction is as accurate for Merton (4.5 × 10−7) and about four times less accurate for Kou (1.9 × 10−6).
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 36 related papers). Coverage is partial, so this is neither replication nor refutation.
Extracted 2026-11-17 · same card as JSON · A language model chose these items: check the source sentence before relying on one.
Cited References (57)
- Physics-informed neural network for option pricing weather derivatives model (2025) — link
- Application of physics informed neural networks to partial integro-differential equations in financial modeling and decision making (2026)
- Deep splitting method for parabolic PDE s (2021)
- Option valuation using the fast F ourier transform (1999)
- Numerical solving of the generalized B lack– S choles differential equation using L aguerre neural network (2021) — link
- Bridging traditional and machine learning-based algorithms for solving PDE s: the random feature method (2022)
- A finite difference scheme for option pricing in jump diffusion and exponential L 'evy models (2005)
- Approximation by superpositions of a sigmoidal function (1989)
and 49 more
- Error analysis for physics-informed neural networks ( PINN s) approximating K olmogorov PDE s (2022) — link
- Local extreme learning machines and domain decomposition for solving linear and nonlinear partial differential equations (2021)
- Physics informed extreme learning machine ( PIELM ) – a rapid method for the numerical solution of partial differential equations (2020)
- A novel pricing method for E uropean options based on F ourier-cosine series expansions (2008)
- On the use of singular functions with finite element approximations (1973)
- Partial Differential Equations of Parabolic Type (1964)
- The differentiation of pseudo-inverses and nonlinear least squares problems whose variables separate (1973)
- Random feature neural networks learn B lack– S choles type PDE s without curse of dimensionality (2023)
- Managing smile risk (2002)
- Option pricing in the H eston model with physics inspired neural networks (2024) — link
- Solving high-dimensional partial differential equations using deep learning (2018)
- A closed-form solution for options with stochastic volatility with applications to bond and currency options (1993)
- A proof that rectified deep neural networks overcome the curse of dimensionality in the numerical approximation of semilinear heat equations (2020) — link
- ADI finite difference schemes for option pricing in the H eston model with correlation (2010)
- Random neural networks for rough volatility (2023)
- A jump-diffusion model for option pricing (2002)
- Multilayer feedforward networks with a nonpolynomial activation function can approximate any function (1993)
- Option pricing under a H eston volatility model using ADI schemes (2015)
- Fast deterministic pricing of options on L 'evy driven assets (2004) — link
- Option pricing when underlying stock returns are discontinuous (1976)
- A finite element method for crack growth without remeshing (1999)
- Differential deep learning and physics-informed neural networks for fast derivative pricing (2025)
- Full error analysis of the random deep splitting method for nonlinear parabolic PDE s and PIDE s (2025)
- Universal approximation property of B anach space-valued random feature models including random neural networks (2023)
- Convergence remedies for non-smooth payoffs in option pricing (2003)
- Weighted sums of random kitchen sinks: Replacing minimization with randomization in learning (2008)
- Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations (2019)
- Implicit neural representations with periodic activation functions (2020)
- Towards fast option pricing PDE solvers powered by PIELM (2025)
- BENCHOP–SLV : the BENCH marking project in O ption P ricing – S tochastic and L ocal V olatility problems (2019) — link
- An extreme learning machine-based method for computational PDE s in higher dimensions (2024)
- A splitting strategy for the calibration of jump-diffusion models (2020) — link
- Option pricing and local volatility surface by physics-informed neural network (2024)
- DeepSVM : Learning stochastic volatility models with physics-informed deep operator networks — → 2512.07162
- The application of the inverse physics-informed neural network in financial calibration tasks (2026)
- The B lack– S choles equation in stochastic volatility models (2010)
- A table of normal integrals (1980)
- Stochastic Calculus for Finance II : Continuous-Time Models (2004)
- Financial Modelling with Jump Processes (2004)
- A subspace, interior, and conjugate gradient method for large-scale bound-constrained minimization problems (1999)
- DGM : A deep learning algorithm for solving partial differential equations (2018)
- Deep learning volatility: a deep neural network perspective on pricing and calibration in (rough) volatility models (2021)
- Robust numerical methods for contingent claims under jump diffusion processes (2005)
- Statistical inverse problems: discretization, model reduction and inverse crimes (2007)
- Can physics-informed neural networks beat the finite element method? (2024) — link
- Weak baselines and reporting biases lead to overoptimism in machine learning for fluid-related partial differential equations (2024) — link
- Option Valuation under Stochastic Volatility (2000)
- Time dependent H eston model (2010) — link
- Monte C arlo Methods in Financial Engineering (2004)
1 of 57 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.