Paper: arXiv 2405.02570

Authors: Jiefei Yang, Guanglian Li

Abstract

We propose an efficient and easy-to-implement gradient-enhanced least squares Monte Carlo method for computing price and Greeks (i.e., derivatives of the price function) of high-dimensional American options. It employs the sparse Hermite polynomial expansion as a surrogate model for the continuation value function, and essentially exploits the fast evaluation of gradients. The expansion coefficients are computed by solving a linear least squares problem that is enhanced by gradient information of simulated paths. We analyze the convergence of the proposed method, and establish an error estimate in terms of the best approximation error in the weighted \(H^1\) space, the statistical error of solving discrete least squares problems, and the time step size. We present comprehensive numerical experiments to illustrate the performance of the proposed method. The results show that it outperforms the state-of-the-art least squares Monte Carlo method with more accurate price, Greeks, and optimal exercise strategies in high dimensions but with nearly identical computational cost, and it can deliver comparable results with recent neural network-based methods up to dimension 100.

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 involves advanced mathematics including weighted Sobolev spaces, backward stochastic differential equations, and Malliavin calculus for convergence analysis, while also providing comprehensive numerical experiments with performance comparisons and error estimates.

Research Flowchart

  flowchart TD
  Start(["Research Goal<br>Efficient Pricing & Greeks for<br>High-Dim American Options"]) --> Inputs

  subgraph Inputs ["Data/Inputs"]
      S(["Simulated Asset Paths"])
      H(["Hermite Polynomials"])
      O(["American Option Payoffs"])
  end

  Inputs --> Methodology

  subgraph Methodology ["Methodology Steps"]
      G(["Compute Gradient Information<br>from Paths"])
      LS(["Solve Gradient-Enhanced<br>Least Squares Problem"])
      C(["Construct Sparse Hermite<br>Surrogate Model"])
  end

  Methodology --> Processes

  subgraph Processes ["Computational Processes"]
      P1(["Price via Backward Induction<br>using Surrogate Model"])
      P2(["Compute Greeks<br>Numerically"])
      P3(["Determine Optimal<br>Exercise Strategy"])
  end

  Processes --> Outcomes

  subgraph Outcomes ["Key Findings/Outcomes"]
      F1(["High Accuracy in Prices & Greeks"])
      F2(["Efficient High-Dim Scaling<br>up to d=100"])
      F3(["Outperforms Standard LS-MC<br>and Neural Networks"])
  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 proposes a gradient-enhanced least squares Monte Carlo (G-LSM) method that uses sparse Hermite polynomial expansions to price and hedge high‑dimensional American options.

Source sentence
We propose an efficient and easy-to-implement gradient-enhanced least squares Monte Carlo method for computing price and Greeks (i.e., derivatives of the price function) of high-dimensional American options.

Assumes 3

  • The discounted payoff function g_k(·) is L‑Lipschitz continuous. model
    Source sentence
    Assumption 5.1. The discounted payofffunction gk(·) is L-Lipschitz continuous: for k = 0, 1, . . . , N (5.1) |gk(s) −gk(s′)| ≤L∥s −s′∥, ∀s, s′ ∈Rd.
  • Underlying assets follow the multi‑asset Black‑Scholes model with correlated geometric Brownian motions. market
    Source sentence
    One of the most classical models for high-dimensional American option pricing is the multi-asset Black‑Scholes model. Under the risk‑neutral probability Q, the prices of d underlying assets, St = S1t , . . . , Sdt ⊤, follow the correlated geometric Brownian motions (2.4) dSi t = (r −δi)Si t dt + σiSi t d ˜W i t with Si
  • The continuation value function can be reformulated as a function of independent Brownian motions after a linear transformation. model
    Source sentence
    Since the corresponding CVF can be reformulated as a function of independent Brownian motions and the associated best approximation error in the sparse Hermite polynomial ansatz space PI,k can be rigorously established, cf. Lemma 5.2.

Data

  • Bermudan geometric basket put (benchmark) benchmark unknown
    Source sentence
    Geometric basket options are benchmark tests for high-dimensional option pricing problems, since they can be reduced to one-dimensional problems, and thus highly accurate prices are available.
  • Max‑call with d symmetric assets synthetic unknown
    Source sentence
    Max-call with d symmetric assets K = 100, T = 3, r = 0.05, δi = 0.1, σi = 0.2, ρij = 0, N = 9
  • Max‑call with d asymmetric assets synthetic unknown
    Source sentence
    Max-call with d asymmetric assets σi = ( 0.08 + 0.32 × (i −1)/(d −1), if d ≤5 0.1 + i/(2d), if d > 5
  • Put option under Heston model synthetic unknown
    Source sentence
    Put option under Heston model K = 10, T = 0.25, r = 0.1, v0 = 0.0625, ρ = 0.1, κ = 5, θ = 0.16, ν = 0.9, N = 50

Costs

Not applicable The paper reports no trading or execution result.

Finds 3

  • G‑LSM achieves substantially lower relative pricing errors than standard LSM in high dimensions (e.g., 0.55% vs 5.15% for d=15). performance
    Source sentence
    G-LSM has a relative error 0.55% for d = 15, which is almost ten times smaller than that by LSM.
  • G‑LSM matches the accuracy of recent deep‑neural‑network methods up to dimension 100 while keeping computational cost nearly identical to LSM. performance
    Source sentence
    It outperforms the state-of-the-art least squares Monte Carlo method with more accurate price, Greeks, and optimal exercise strategies in high dimensions but with nearly identical computational cost, and it can deliver comparable results with recent neural network-based methods up to dimension 100.
  • The method retains low complexity for gradient evaluation and ease of implementation, making it suitable for high‑dimensional problems. theoretical
    Source sentence
    The method enjoys low complexity for the gradient evaluation, ease of implementation and high accuracy for high-dimensional problems.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 39 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 1 paper in the archive (1 by independent authors)

Code

https://github.com/jiefeiy/glsm-american — 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 (27)

  1. Pricing and hedging American-style options: a simple simulation-based approach (2010) — link
  2. Monte-Carlo Valuation of American Options: Facts and New Algorithms to Improve Existing Methods (2012) — link
  3. Error expansion for the discretization of backward stochastic differential equations (2006) — link
  4. A regression-based Monte Carlo method to solve backward stochastic differential equations (2005) — link
  5. Pricing Multi-Asset American Options: A Finite Element Method-of-Lines with Smooth Penalty (2007) — link
  6. Reflected solutions of backward SDE’s, and related obstacle problems for PDE’s (1997) — link
  7. A Fourier-Based Valuation Method for Bermudan and Barrier Options under Heston’s Model (2011) — link
  8. Valuing American Options by Simulation: A Simple Least-Squares Approach (2001) — link
and 19 more
  1. Regression methods for pricing complex American-style options (2001) — link
  2. Stochastic Differential Equations: An Introduction with Applications. (1987) — link
  3. Methods of conjugate gradients for solving linear systems (1952) — link
  4. Kriging metamodels and experimental design for Bermudan option pricing (2018) — link
  5. Deep Learning-Based Numerical Methods for High-Dimensional Parabolic Partial Differential Equations and Backward Stochastic Differential Equations (2017) — link
  6. DGM: A deep learning algorithm for solving partial differential equations (2018) — link
  7. Error analysis of the Wiener–Askey polynomial chaos with hyperbolic cross approximation and its application to differential equations with random input (2017) — link
  8. Deep backward schemes for high-dimensional nonlinear PDEs (2019) — link
  9. Pricing and Hedging American-Style Options with Deep Learning (2020) — link
  10. Sparse Polynomial Chaos Expansions: Literature Survey and Benchmark (2021) — link
  11. Deep neural network framework based on backward stochastic differential equations for pricing and hedging American options in high dimensions (2020) — link
  12. Pricing High-Dimensional Bermudan Options with Hierarchical Tensor Formats (2023) — link
  13. Backward Stochastic Differential Equations (2017) — link
  14. Tools for Computational Finance (2004) — link
  15. Deep stochastic optimization in finance (2022) — link
  16. Efficient pricing and hedging of high-dimensional American options using deep recurrent networks (2023) — link
  17. Convergence of the Backward Deep BSDE Method with Applications to Optimal Stopping Problems (2023) — link
  18. Neural optimal stopping boundary (2024) — link
  19. On sparse grid interpolation for American option pricing with multiple underlying assets — → 2309.08287

1 of 27 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.