Paper: arXiv 2401.06740
Authors: Emmanuil H. Georgoulis, Antonis Papapantoleon, Costas Smaragdakis
Abstract
We develop a novel deep learning approach for pricing European basket options written on assets that follow jump-diffusion dynamics. The option pricing problem is formulated as a partial integro-differential equation, which is approximated via a new implicit-explicit minimizing movement time-stepping approach, involving approximation by deep, residual-type Artificial Neural Networks (ANNs) for each time step. The integral operator is discretized via two different approaches: (a) a sparse-grid Gauss-Hermite approximation following localised coordinate axes arising from singular value decompositions, and (b) an ANN-based high-dimensional special-purpose quadrature rule. Crucially, the proposed ANN is constructed to ensure the appropriate asymptotic behavior of the solution for large values of the underlyings and also leads to consistent outputs with respect to a priori known qualitative properties of the solution. The performance and robustness with respect to the dimension of these methods are assessed in a series of numerical experiments involving the Merton jump-diffusion model, while a comparison with the deep Galerkin method and the deep BSDE solver with jumps further supports the merits of the proposed approach.
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: The paper presents advanced mathematical theory involving minimizing movement methods, implicit-explicit schemes, and PIDEs, indicating high math complexity, while the numerical experiments and comparison to benchmarks provide significant empirical validation but lack full backtest-ready details.
Research Flowchart
flowchart TD
A["Research Goal<br>Pricing European Basket Options<br>in Jump-Diffusion Models"] --> B["Formulate Problem<br>Partial Integro-Differential Equation"]
B --> C["Deep Implicit-Explicit<br>Minimizing Movement Method"]
C --> D{"Integral Operator<br>Discretization"}
D --> E["Sparse-Grid Gauss-Hermite<br>SVD Approximation"]
D --> F["ANN-based<br>High-Dim Quadrature Rule"]
E & F --> G["Deep Residual ANN Training<br>with Asymptotic Constraints"]
G --> H{"Performance Evaluation"}
H --> I["Experiments on<br>Merton Jump-Diffusion Model"]
H --> J["Comparison with<br>Deep Galerkin & Deep BSDE"]
I & J --> K["Key Findings<br>Effective High-Dim Pricing<br>Robustness & Consistency Verified"]
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 novel deep learning method using an implicit‑explicit minimizing‑movement time‑stepping scheme with deep residual ANNs to price European basket options under jump‑diffusion dynamics.
Source sentence
We develop a novel deep learning approach for pricing European basket options written on assets that follow jump-diffusion dynamics. The option pricing problem is formulated as a partial integro-differential equation, which is approximated via a new implicit-explicit minimizing movement time-stepping approach, involvin
Assumes 5
- Asset prices follow the Merton jump‑diffusion model, i.e., a geometric Brownian motion plus a Poisson jump component. model
Source sentence
In the Merton model, asset prices follow a superposition of a geometric Brownian motion and a Poisson process with randomly distributed jumps
- The diffusion and jump coefficients a_{ij}, b_j belong to the Sobolev space W^{1,∞}([0,∞)^d). model
Source sentence
it is sufficient to assume that aij, bj ∈W 1,∞([0, ∞)^{d}), using standard Sobolev space notation.
- Homogeneous Dirichlet boundary conditions are assumed on the truncated domain for simplicity. model
Source sentence
In the above discussion, we assumed, for simplicity, homogeneous Dirichlet boundary conditions on the domain of the truncated boundary ∂Ω.
- Large volatilities and asset correlations are deliberately chosen to test the method in challenging scenarios. model
Source sentence
Let us point out that we have intentionally selected large values for the volatilities and asset correlations, in order to test the performance of the method in challenging scenarios.
Show 1 more
- Domain truncation is applied, projecting option prices onto a bounded subset of ℝ^d_+ and approximating extreme moneyness values. other
Source sentence
A domain truncation method is introduced to enhance the accuracy of the numerical schemes. This involves projecting the option prices onto a bounded subset of
Data
- Synthetic Merton jump‑diffusion model (5 and 15 assets) synthetic unknown
Source sentence
Below, we present numerical experiments for dimensions d = 5 and d = 15.
Costs
Not applicable
Finds 5
- Both the implicit‑explicit Euler and BDF‑2 schemes achieve absolute errors on the order of 10⁻³ compared to Quasi‑Monte Carlo, with the Euler scheme slightly more accurate for deep in‑the‑money options. performance
Source sentence
We can observe that both methods work very well compared to the QMC simulation, and the difference to the Quasi Monte Carlo is only visible for deep in-the-money options, with moneyness level above 2. Indeed, the absolute error is on the order of 10−3.
- The implicit‑explicit Euler method is about 1.8 times slower than the BDF‑2 scheme due to its smaller time step. performance
Source sentence
The implicit-explicit Euler method appears slightly more accurate than the BDF-2 scheme, due to the smallest time-stepping. This comes at a computational cost, since it is approximately 1.8 times slower.
- Using the ANN‑based quadrature for the integral operator also yields absolute errors around 10⁻³, confirming its effectiveness. performance
Source sentence
We can observe that the ANN computation of the integral operator also works very well, and the absolute error remains in the order of 10−3.
- The deep IMEX method is competitive with MC and deep‑BSDE, being less than twice as expensive as MC and less than three times as expensive as deep‑BSDE for similar accuracy. performance
Source sentence
deep IMEX approach is competitive with both MC and deep BSDE methodologies in terms of error and runtime, (less than twice as expensive as MC and less then three times as expensive as deep BSDE, for similar accuracy)
- Deep IMEX provides a complete space‑time solution, enabling straightforward computation of Greeks, and is more economical in runtime and parameter count than DGM. theoretical
Source sentence
As such, it is straightforward to infer further quantities, such as Greeks, from the computed space-time solution.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 31 related papers). Coverage is partial, so this is neither replication nor refutation.
4 papers in the archive : Error Analysis of Deep PDE Solvers for Option Pricing
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 (16)
- Jumps and Stochastic Volatility: Exchange Rate Processes Implicit in D eutsche Mark Options (1996)
- Option Valuation Using the Fast F ourier Transform (1999)
- Optimal Damping with Hierarchical Adaptive Quadrature for Efficient F ourier Pricing of Multi-Asset Options in L évy Models (2023)
- Quasi- M onte C arlo for Efficient F ourier Pricing of Multi-Asset Options (2024)
- A Jump-Diffusion Model for Option Pricing (2002)
- Option Pricing When Underlying Stock Returns are Discontinuous (1976)
- Jump-Diffusion Models (2003)
- DGM : A deep learning algorithm for solving partial differential equations (2018)
and 8 more
- Discrete gradient flow approximations of high dimensional evolution partial differential equations via deep neural networks (2023)
- A DEEP SOLVER FOR BSDE S WITH JUMPS (2022)
- The Deep R itz Method: A Deep Learning-Based Numerical Algorithm for Solving Variational Problems (2018)
- Deep N itsche Method: Deep R itz Method with Essential Boundary Conditions (2019)
- Implicit-explicit multistep methods for quasilinear parabolic equations (1999)
- D elft B lue S upercomputer ( P hase 1) (2022)
- On the distribution of points in a cube and the approximate evaluation of integrals (1967)
- Scrambling Sobol’ and Niederreiter–Xing Points (1998)
0 of 16 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.