Paper: arXiv 2405.11392

Authors: Yunfei Peng, Pengyu Wei, Wei Wei

Abstract

We propose a deep learning algorithm for high dimensional optimal stopping problems. Our method is inspired by the penalty method for solving free boundary PDEs. Within our approach, the penalized PDE is approximated using the Deep BSDE framework proposed by \cite{“weinan2017deep”}, which leads us to coin the term “Deep Penalty Method (DPM)” to refer to our algorithm. We show that the error of the DPM can be bounded by the loss function and \(O(\frac{“1”}λ)+O(λh) +O(\sqrt{“h”})\), where \(h\) is the step size in time and \(λ\) is the penalty parameter. This finding emphasizes the need for careful consideration when selecting the penalization parameter and suggests that the discretization error converges at a rate of order \(\frac{“1”}{“2”}\). We validate the efficacy of the DPM through numerical tests conducted on a high-dimensional optimal stopping model in the area of American option pricing. The numerical tests confirm both the accuracy and the computational efficiency of our proposed algorithm.

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 presents advanced mathematical formulations including stochastic differential equations, PDE analysis, and error bounds, scoring high on math complexity. It provides empirical validation through numerical tests on American option pricing up to 100 dimensions, demonstrating practical computational efficiency, thus scoring moderately high on empirical rigor.

Research Flowchart

  flowchart TD
  Start["Research Goal:<br>Solve High-Dim Optimal Stopping Problems"] --> Method["Methodology:<br>Deep Penalty Method DPM"]
  Method --> Input["Inputs:<br>High-Dim American Option Model"]
  Input --> Process["Process:<br>Deep BSDE & Penalized PDE Solver"]
  Process --> Analyze["Analysis:<br>Error Bound: 1/λ + λh + √h"]
  Analyze --> Outcome["Outcomes:<br>Accurate & Computationally Efficient"]

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 deep learning algorithm called the Deep Penalty Method (DPM) for solving high dimensional optimal stopping problems.

Source sentence
We propose a deep learning algorithm for high dimensional optimal stopping problems.

Assumes 2

  • b and σ are uniformly Hölder continuous in time with exponent 1/2 and uniformly Lipschitz continuous in space. model
    Source sentence
    Assumption 1 b and σ are uniformly H¨older continuous in t with exponent 1/2 and uniformly Lipschitz continuous in x, i.e., there exists an L > 0 such that, |ϕ(t, x) −ϕ(s, y)| ≤L(|t −s|^{1/2} + |x −y|)
  • The differential operator L is uniformly parabolic, i.e., there exists λ1>0 such that aij(t,x)ξiξj ≥ λ1|ξ|^2 for all (t,x). model
    Source sentence
    Suppose that L is uniformly parabolic for any T > 0, i.e., there exists a positive constant λ1 > 0, such that for any ξ = (ξ1, ξ2, . . . , ξd)T ∈Rd

Data

  • high-dimensional American index put option synthetic unknown
    Source sentence
    Finally, we test our algorithm on a high-dimensional American index put option. This pricing problem for this type of option can be simplified to a standard one-dimensional option pricing case, and thus providing a good benchmark for our analysis.

Costs

Not applicable The paper reports no trading or execution result.

Finds 2

  • The error of the DPM can be bounded by the cost function and O(1/λ)+O(λh)+O(√h), with a convergence rate of O(√h) when λ=1/√h. theoretical
    Source sentence
    We show that the error of the DPM can be bounded by the cost function and O( 1/λ) + O(λh) + O( √h), where h is the step size in time and λ
  • Numerical tests confirm that the DPM achieves relative errors significantly below 1% across dimensions up to d=200, with loss variance between O(10⁻⁸) and O(10⁻⁷). performance
    Source sentence
    ... The data reveals that the DPM maintains a relative error significantly below 1% across all tested cases. Furthermore, the consistently low loss variance (O(10−8) −O(10−7)) across all high-dimensional tests underscores the stability of the DPM during the final stages of optimization.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 29 related papers). Coverage is partial, so this is neither replication nor refutation.

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

Code

https://github.com/weiw1422/DeepPenaltyMethod — the authors' own repository

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.

Cited References (39)

  1. Optimal Stochastic Control, Stochastic Target Problems, and Backward SDE (2012) — link
  2. Numerical Solution of Stochastic Differential Equations with Jumps in Finance (2010) — link
  3. Continuous-time Stochastic Control and Optimization with Financial Applications (2009) — link
  4. Adam: A Method for Stochastic Optimization (2014) — link
  5. Options, Futures and Other Derivatives (2003) — link
  6. A stochastic mesh method for pricing high-dimensional American options (2004) — link
  7. A Penalty Method for the Numerical Solution of Hamilton–Jacobi–Bellman (HJB) Equations in Finance (2011) — link
  8. Stochastic Control Representations for Penalized Backward Stochastic Differential Equations (2015) — link
and 31 more
  1. PRICING OF HIGH-DIMENSIONAL AMERICAN OPTIONS BY NEURAL NETWORKS (2010) — link
  2. Funding liquidity, debt tenor structure, and creditor’s belief: an exogenous dynamic debt run model (2015) — link
  3. Backward stochastic differential equations and quasilinear parabolic partial differential equations (2005) — link
  4. The Effect of Nonsmooth Payoffs on the Penalty Approximation of American Options (2013) — link
  5. Discrete-time approximation for continuously and discretely reflected BSDEs (2008) — link
  6. On the rate of convergence of the binomial tree scheme for American options (2007) — link
  7. Quadratic Convergence for Valuing American Options Using a Penalty Method (2002) — link
  8. On the Use of Policy Iteration as an Easy Way of Pricing American Options (2012) — link
  9. Forward-backward stochastic differential equations and quasilinear parabolic PDEs (1999) — link
  10. Valuing American Options by Simulation: A Simple Least-Squares Approach (2001) — link
  11. Optimal Switching at Poisson Random Intervention Times (2013) — link
  12. Artificial neural networks for solving ordinary and partial differential equations (1998) — link
  13. Discrete-time approximation and Monte-Carlo simulation of backward stochastic differential equations (2004) — link
  14. Deep Learning-Based Numerical Methods for High-Dimensional Parabolic Partial Differential Equations and Backward Stochastic Differential Equations (2017) — link
  15. DGM: A deep learning algorithm for solving partial differential equations (2018) — link
  16. Machine Learning Approximation Algorithms for High-Dimensional Fully Nonlinear Partial Differential Equations and Second-order Backward Stochastic Differential Equations (2019) — link
  17. Asymptotic Expansion as Prior Knowledge in Deep Learning Method for High Dimensional BSDEs (2017) — link
  18. Forward–Backward Stochastic Neural Networks: Deep Learning of High-Dimensional Partial Differential Equations (2023) — link
  19. Deep optimal stopping (2018) — link
  20. Solving high-dimensional partial differential equations using deep learning (2018) — link
  21. Machine Learning for Semi Linear PDEs (2019) — link
  22. Convergence of the deep BSDE method for coupled FBSDEs (2020) — link
  23. Deep backward schemes for high-dimensional nonlinear PDEs (2019)
  24. Optimal switching at Poisson random intervention times (2016) — link
  25. Deep backward schemes for high-dimensional nonlinear PDEs (2019) — link
  26. Deep neural network framework based on backward stochastic differential equations for pricing and hedging American options in high dimensions (2020) — link
  27. Pricing American Options: A Duality Approach (2004) — link
  28. Convergence of Binomial Tree Methods for European/American Path-Dependent Options (2004) — link
  29. Credit Valuation Adjustment with Replacement Closeout: Theory and Algorithms (2022) — link
  30. Brownian Motion and Stochastic Calculus. (1989) — link
  31. Efficient pricing and hedging of high-dimensional American options using deep recurrent networks (2023) — link

0 of 39 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.