Paper: arXiv 2311.07231

Authors: Rawin Assabumrungrat, Kentaro Minami, Masanori Hirano

Abstract

Option pricing, a fundamental problem in finance, often requires solving non-linear partial differential equations (PDEs). When dealing with multi-asset options, such as rainbow options, these PDEs become high-dimensional, leading to challenges posed by the curse of dimensionality. While deep learning-based PDE solvers have recently emerged as scalable solutions to this high-dimensional problem, their empirical and quantitative accuracy remains not well-understood, hindering their real-world applicability. In this study, we aimed to offer actionable insights into the utility of Deep PDE solvers for practical option pricing implementation. Through comparative experiments, we assessed the empirical performance of these solvers in high-dimensional contexts. Our investigation identified three primary sources of errors in Deep PDE solvers: (i) errors inherent in the specifications of the target option and underlying assets, (ii) errors originating from the asset model simulation methods, and (iii) errors stemming from the neural network training. Through ablation studies, we evaluated the individual impact of each error source. Our results indicate that the Deep BSDE method (DBSDE) is superior in performance and exhibits robustness against variations in option specifications. In contrast, some other methods are overly sensitive to option specifications, such as time to expiration. We also find that the performance of these methods improves inversely proportional to the square root of batch size and the number of time steps. This observation can aid in estimating computational resources for achieving desired accuracies with Deep PDE solvers.

Complexity vs Empirical Score

  • Math Complexity: 7.5/10
  • Empirical Rigor: 8.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs advanced high-dimensional PDE/BSDE theory and deep learning optimization, indicating high mathematical complexity, while its rigorous comparative experiments, ablation studies, and quantitative error scaling analysis provide strong empirical evidence suitable for implementation.

Research Flowchart

  flowchart TD
  Start(["Research Goal: Assess Deep PDE Solvers for<br>Option Pricing in High Dimensions"]) --> Inputs
  
  Inputs["Data/Inputs:<br>Multi-Asset Options e.g., Rainbow Options<br>Asset Model Simulations"] --> Methodology
  
  Methodology["Methodology:<br>Comparative Experiments &<br>Ablation Studies"] --> Processes
  
  Processes["Computational Processes:<br>1. Deep PDE Solvers e.g., DBSDE<br>2. Neural Network Training<br>3. Model Evaluation"] --> Findings
  
  Findings["Key Findings/Outcomes:<br>1. Identified 3 Error Sources<br>2. DBSDE is most robust & accurate<br>3. Accuracy improves ∝ 1/√Batch Size & Time Steps"]

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 provides practical guidelines and empirical evaluation of deep PDE/BSDE solvers for high‑dimensional option pricing, highlighting error sources and recommending configurations.

Source sentence
Our primary objective is to provide practical guidelines for the design of models and training algorithms in option pricing using Deep PDE/BSDE solvers.

Assumes 4

  • Option pricing is performed on a high‑dimensional rainbow (best‑of) call option with 20 underlying assets. model
    Source sentence
    Our standard setting uses a call, best-of option of N = 20 underlying assets.
  • Asset price dynamics follow the specified stochastic model with parameters ρ, κ, ξ, etc., satisfying 2κθ > ξ² to keep variance positive. model
    Source sentence
    The values of ρ, κ, and ξ are the default parameters of options provided by [25]
  • Neural network architectures and hyper‑parameters are kept constant across all tested algorithms. execution
    Source sentence
    In addition, regarding the standard hyperparameters applied across all tested algorithms, the neural networks have a total of 4…
  • Batch size can be increased arbitrarily (e.g., using GPUs) without affecting computation feasibility. execution
    Source sentence
    Since there are confirmed advantages to enlarging the batch, it is strongly recommended that batch size in any algorithm is as large as possible. This is especially advised in environments equipped with GPUs where it is unlikely that a larger batch size will hold back computation.

Data

  • Synthetic asset price simulations for high‑dimensional rainbow options synthetic unknown
    20 underlying assets
    Source sentence
    Our standard setting uses a call, best-of option of N = 20 underlying assets.

Costs

Not applicable The paper reports no trading or execution result.

Finds 4

  • The Deep BSDE method consistently outperforms other deep PDE solvers and is robust to variations in option parameters such as time to expiration. performance
    Source sentence
    Our results indicate that the Deep BSDE method (DBSDE) is superior in performance and exhibits robustness against variations in option specifications.
  • Error decreases proportionally to 1/√batch size and 1/√time steps, aiding computational budgeting for target accuracies. performance
    Source sentence
    to the square root of the batch size (proportion to 1/√batch size) and the number of time discretization steps (proportion to 1/√time steps).
  • Small batch sizes may yield inconsistent price estimates, recommending the use of the largest feasible batch. negative
    Source sentence
    a small batch size might not fairly reflect the price distribution, which could result in inconsistent estimated option prices.
  • A strong linear relationship exists between 1/√time steps and error, confirming that finer time discretization reduces pricing error, especially for MDBDP. statistical
    Source sentence
    There exists a particularly strong linear relationship between 1/√time steps and error.

Checked by others

Supported by

  • Deep Quadratic Hedging (2022)
    On the question: How does increasing the number of time steps affect pricing error in deep quadratic hedging and deep PDE solvers?
    Compare the findings

    This paper: A strong linear relationship exists between 1/√time steps and error, confirming that finer time discretization reduces pricing error, especially for MDBDP.

    That paper: Increasing the number of time steps reduces the mean squared error of the option price, though errors accumulate over time.

    Both find finer discretization lowers error: A reports reduced MSE with more steps (though errors accumulate), B shows error ∝1/√steps, especially for MDBDP.

Contradicted by

  • Monte-Carlo Option Pricing in Quantum Parallel (2025)
    On the question: How does the number of time steps affect option pricing error in numerical methods?
    Compare the findings

    This paper: A strong linear relationship exists between 1/√time steps and error, confirming that finer time discretization reduces pricing error, especially for MDBDP.

    That paper: Mean pricing error shows little dependence on the number of time steps beyond about 100 steps, where error plateaus.

    A finds error ∝1/√steps (linear with 1/√time steps), B finds error plateaus after ~100 steps, showing little dependence beyond that.

Cited by 2 papers in the archive (1 by independent authors)

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 (45)

  1. An overview on deep learning-based approximation methods for partial differential equations (2020) — link
  2. Neural networks-based algorithms for stochastic control and PDE s in finance (2021) — link
  3. Multilevel Picard iterations for solving smooth semilinear parabolic heat equations (2021)
  4. On multilevel Picard numerical approximations for high-dimensional nonlinear parabolic partial differential equations and high-dimensional nonlinear backward stochastic differential equations (2019)
  5. Overcoming the curse of dimensionality in the numerical approximation of semilinear parabolic partial differential equations (2020)
  6. Deep Learning-Based Numerical Methods for High-Dimensional Parabolic Partial Differential Equations and Backward Stochastic Differential Equations (2017)
  7. Solving high-dimensional partial differential equations using deep learning (2018)
  8. Solving parametric PDE problems with artificial neural networks (2021)
and 37 more
  1. DGM : A deep learning algorithm for solving partial differential equations (2018)
  2. Machine Learning Approximation Algorithms for High-Dimensional Fully Nonlinear Partial Differential Equations and Second-order Backward Stochastic Differential Equations (2019)
  3. The Deep Ritz Method: A Deep Learning-Based Numerical Algorithm for Solving Variational Problems (2018)
  4. Asymptotic Expansion as Prior Knowledge in Deep Learning Method for High dimensional BSDE s (2019)
  5. Solving the Kolmogorov PDE by Means of Deep Learning (2021)
  6. Deep Splitting Method for Parabolic PDE s (2021)
  7. Unbiased Deep Solvers for Linear Parametric PDE s (2021)
  8. Machine Learning for Semi Linear PDE s (2019)
  9. Deep Primal-Dual Algorithm for BSDE s: Applications of Machine Learning to CVA and IM (2017) — link
  10. Deep backward schemes for high-dimensional nonlinear PDE s (2020)
  11. Solving high-dimensional Hamilton–Jacobi–Bellman PDE s using neural networks: perspectives from the theory of controlled diffusions and measures on path space (2021)
  12. Three Algorithms for Solving High-Dimensional Fully Coupled FBSDE s Through Deep Learning (2020)
  13. Deep neural network framework based on backward stochastic differential equations for pricing and hedging American options in high dimensions (2021)
  14. Deep Curve-dependent PDE s for affine rough volatility (2019) — link
  15. Numerical simulations for full history recursive multilevel Picard approximations for systems of high-dimensional partial differential equations (2020) — link
  16. An acceleration scheme for deep learning-based BSDE solver using weak expansions (2020)
  17. Algorithms for solving high dimensional PDEs: from nonlinear Monte Carlo to machine learning (2021)
  18. Backward stochastic differential equations and quasilinear parabolic partial differential equations (1992)
  19. Three ways to solve partial differential equations with neural networks — A review (2021)
  20. Approximation Error Analysis of Some Deep Backward Schemes for Nonlinear PDEs (2022)
  21. The pricing of options and corporate liabilities (1973)
  22. A proof that rectified deep neural networks overcome the curse of dimensionality in the numerical approximation of semilinear heat equations (2020)
  23. A proof that artificial neural networks overcome the curse of dimensionality in the numerical approximation of Black-Scholes partial differential equations (2020)
  24. Partial differential equation representations of derivatives with bilateral counterparty risk and funding costs (2011)
  25. Funding strategies, funding costs (2013)
  26. Implied Index And Option Pricing Errors: Evidence From The Taiwan Option Market (2011)
  27. A Closed-form Pricing Solution for Options on Assets with Pricing Errors (2021) — link
  28. Paul Wilmott on Quantitative Finance (2006)
  29. Partial Differential Equations for Option Pricing (2009) — link
  30. Degree of Mispricing with the Black-Scholes Model and Nonparametric Cures (2003) — link
  31. Pricing European Options with the Black-Scholes and Monte Carlo Methods: a Comparative Study (2022) — link
  32. Monte Carlo simulations for complex option pricing (2011) — link
  33. Monte Carlo Estimation of Project Volatility for Real Options Analysis (2006) — link
  34. Risk-neutral valuation: Pricing and hedging of financial derivatives (2004)
  35. Stochastic Calculus for Finance II: Continuous-Time Models (2004)
  36. A Closed-Form Solution for Options with Stochastic Volatility with Applications to Bond and Currency Options (2015) — link
  37. Numerical methods for backward stochastic differential equations: A survey (2023) — link

0 of 45 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.