Paper: arXiv 2507.17606

Authors: Jasper Rou

Abstract

In this research, we explore neural network-based methods for pricing multidimensional American put options under the BlackScholes and Heston model, extending up to five dimensions. We focus on two approaches: the Time Deep Gradient Flow (TDGF) method and the Deep Galerkin Method (DGM). We extend the TDGF method to handle the free-boundary partial differential equation inherent in American options. We carefully design the sampling strategy during training to enhance performance. Both TDGF and DGM achieve high accuracy while outperforming conventional Monte Carlo methods in terms of computational speed. In particular, TDGF tends to be faster during training than DGM.

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 demonstrates high mathematical complexity through advanced PDE theory, operator splitting, and neural network optimization for free-boundary problems, while also showing strong empirical rigor with concrete backtesting results comparing TDGF and DGM against Monte Carlo methods in up to five dimensions under Black-Scholes and Heston models.

Research Flowchart

  flowchart TD
  A["Research Goal: Price Multidim American Put Options"] --> B["Methodology"]
  subgraph B ["Key Methodologies"]
      B1["Time Deep Gradient Flow TDGF"]
      B2["Deep Galerkin Method DGM"]
      B3["Monte Carlo Benchmark"]
  end
  C["Data & Inputs"] --> D["Computational Processes"]
  subgraph C ["Inputs"]
      C1["Black-Scholes & Heston Models<br>1D-5D Configurations"]
      C2["Strategic Sampling Strategy"]
  end
  subgraph D ["Training & Evaluation"]
      D1["Neural Network Training"]
      D2["Free-Boundary PDE Solver"]
  end
  B --> C
  D --> E["Key Findings & Outcomes"]
  subgraph E ["Results"]
      E1["High Pricing Accuracy"]
      E2["TDGF Faster Training than DGM"]
      E3["Outperforms Monte Carlo Speed"]
  end

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 extends the Time Deep Gradient Flow (TDGF) method to handle the free‑boundary PDE for pricing multidimensional American put options and designs a sampling strategy to improve performance.

Source sentence
the TDGF method to handle the free‑boundary PDE inherent in American options.

Assumes 4

  • No dividends are considered and r ≥ 0. market
    Source sentence
    Since we do not consider dividends and assume r ≥0 the best exercise strategy for an American call option is to wait until maturity.
  • Underlying assets follow multidimensional Black‑Scholes dynamics with given correlation matrix. model
    Source sentence
    In the model by Black and Scholes [5], the dynamics of the stock price S follow a geometric Brownian motion. Suppose we have d assets, each following the Black‑Scholes model
  • Underlying assets follow multidimensional Heston stochastic volatility dynamics. model
    Source sentence
    The model by Heston [11] is a popular stochastic volatility model. In d dimensions the dynamics of asset S and variance process V are
  • Training uses a maturity of T = 1 year, K = 100 time steps, and 2000 sampling stages per step. execution
    Source sentence
    We consider a maturity of T = 1.0 year, and take the number of time steps equal to K = 100. We use 2000 sampling stages in each time step.

Data

  • Multidimensional Black‑Scholes model (synthetic) synthetic unknown
    up to five underlying assets
    Source sentence
    We explore neural network‑based methods for pricing multidimensional American put options under the Black–Scholes and Heston model, extending up to five dimensions.
  • Multidimensional Heston model (synthetic) synthetic unknown
    up to five underlying assets
    Source sentence
    We explore neural network‑based methods for pricing multidimensional American put options under the Black–Scholes and Heston model, extending up to five dimensions.

Costs

Not applicable The paper reports no trading or execution result.

Finds 4

  • Both TDGF and DGM achieve high pricing accuracy while significantly outperforming conventional Monte Carlo methods in computational speed. performance
    Source sentence
    Both TDGF and DGM achieve high accuracy while significantly outperforming conventional Monte Carlo methods in terms of computational speed.
  • TDGF tends to be faster during training than DGM. performance
    Source sentence
    Notably, TDGF tends to be faster during training than DGM.
  • Training time (seconds) for an American put option: DGM 0.0015–0.0024, TDGF 0.0017–0.0018 across models. performance
    Source sentence
    TABLE 1. Training time in seconds of the different methods for an American put option in the different models.
  • Computational time (seconds) for pricing: DGM 0.0015–0.0024, TDGF 0.0016–0.0018 across models. performance
    Source sentence
    TABLE 2. Computational time in seconds of the different methods for an American put option in the different models.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 59 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.

Cited References (22)

  1. Deep optimal stopping (2019)
  2. Pricing and hedging A merican-style options with deep learning (2020)
  3. Solving high-dimensional optimal stopping problems using deep learning (2021)
  4. Advanced Simulation-Based Methods for Optimal Stopping and Control: With Applications in Finance (2018)
  5. The pricing of options and corporate liabilities (1973)
  6. Multigrid for A merican option pricing with stochastic volatility (1999)
  7. Option pricing: A simplified approach (1979)
  8. D elft B lue S upercomputer ( P hase 2) (2024)
and 14 more
  1. Discrete gradient flow approximations of high dimensional evolution partial differential equations via deep neural networks (2023)
  2. Optimal stopping via randomized neural networks (2021) — link
  3. A closed-form solution for options with stochastic volatility with applications to bond and currency options (1993)
  4. Computational Methods for Quantitative Finance: Finite Element Methods for Derivative Pricing (2013)
  5. Efficient numerical methods for pricing A merican options under stochastic volatility (2008)
  6. Adam: A method for stochastic optimization (2014) — link
  7. Valuing A merican options by simulation: a simple least-squares approach (2001)
  8. American Options (2005)
  9. The pricing of the A merican option (1992)
  10. Deep learning and A merican options via free boundary framework (2024)
  11. A time-stepping deep gradient flow method for option pricing in (rough) diffusion models — → 2403.00746
  12. Deep penalty methods: A class of deep learning algorithms for solving high dimensional optimal stopping problems — → 2405.11392
  13. DGM : A deep learning algorithm for solving partial differential equations (2018)
  14. An advanced M onte C arlo method for the multi-asset H eston model (2010)

2 of 22 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.