Paper: arXiv 2505.17033

Authors: Maarten van Damme, Rishi Sreedhar, Martin Ganahl

Abstract

Pricing of exotic financial derivatives, such as Asian and multi-asset American basket options, poses significant challenges for standard numerical methods such as binomial trees or Monte Carlo methods. While the former often scales exponentially with the parameters of interest, the latter often requires expensive simulations to obtain sufficient statistical convergence. This work combines the binomial pricing method for options with tensor network techniques, specifically Matrix Product States (MPS), to overcome these challenges. Our proposed methods scale linearly with the parameters of interest and significantly reduce the computational complexity of pricing exotics compared to conventional methods. For Asian options, we present two methods: a tensor train cross approximation-based method for pricing, and a variational pricing method using MPS, which provides a stringent lower bound on option prices. For multi-asset American basket options, we combine the decoupled trees technique with the tensor train cross approximation to efficiently handle baskets of up to \(m = 8\) correlated assets. All approaches scale linearly in the number of discretization steps \(N\) for Asian options, and the number of assets \(m\) for multi-asset options. Our numerical experiments underscore the high potential of tensor network methods as highly efficient simulation and optimization tools for financial engineering.

Complexity vs Empirical Score

  • Math Complexity: 9.0/10
  • Empirical Rigor: 4.0/10
  • Quadrant: Lab Rats — high math complexity, limited empirical testing

Why this score: The paper employs advanced tensor network mathematics (Matrix Product States) and high-dimensional function approximation with bond dimensions, requiring significant theoretical expertise. However, it lacks implementation details, backtests, or statistical metrics, focusing instead on theoretical scaling and potential rather than empirical validation.

Research Flowchart

  flowchart TD
  A["Research Goal: Overcome computational<br>challenges in pricing exotic options"] --> B["Methodology: Tensor Networks<br>(Matrix Product States MPS)"]
  B --> C{"Option Type"}
  C --> D["Asian Options<br>Linear in N steps"]
  C --> E["Multi-Asset American Baskets<br>Linear in m assets"]
  D --> F["Methods: TT Cross Approx &<br>Variational MPS Pricing"]
  E --> G["Method: Decoupled Trees &<br>TT Cross Approx"]
  F & G --> H["Inputs: Market Parameters,<br>Correlation Matrices"]
  H --> I["Computational Process:<br>Linear Scaling O(N) or O(m)"]
  I --> J["Key Outcomes:<br>Efficient Pricing, Accurate Bounds,<br>Scalability up to 8 Assets"]

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 combining binomial option pricing with tensor network techniques (MPS and TTcross) to achieve linear scaling and reduce computational complexity.

Source sentence
Our proposed methods scale linearly with the parameters of interest and significantly reduce the computational complexity of pricing exotics compared to conventional methods.

Assumes 6

  • Assets follow standard geometric Brownian motion. model
    Source sentence
    Following standard practice, we assume that assets follow standard geometric Brownian motion, i.e. given m asset values Si t, i = 1, . . . , m at time t, dSi t = Si t r dt + σi Si t dW i t
  • Risk‑free interest rate is r = 0.1. market
    Source sentence
    A common risk-free interest rate of r = 0.1 was also used.
  • Correlation matrix of assets is positive definite. model
    Source sentence
    this form also ensures that the correlation matrix is positive definite as required by the decoupling trees approach.
  • Derivative lifetime is discretized into N+1 equally spaced time points. model
    Source sentence
    the lifetime T of a derivative is discretized into N+1 discrete, equally spaced time points with separation Δt = T/N.
Show 2 more
  • Binomial up/down parameters u, d, pu, pd are derived from geometric Brownian motion. model
    Source sentence
    The parameters u, d and pu, pd are usually obtained assuming geometric Brownian motion of the asset price St
  • Single parameter is used to model correlation between assets in the decoupled trees approach. model
    Source sentence
    the correlation between assets using a single parameter.

Data

  • Binomial model simulations synthetic unknown
    Source sentence
    Simulations are performed for a binomial model with N = 25, 30, and 50 time steps.
  • Monte Carlo sampling synthetic unknown
    Source sentence
    Monte Carlo results were obtained by running across different number of samples Ns = 10^4 to Ns = 5 × 10^7 samples.

Costs

Not applicable The paper reports no trading or execution result.

Finds 3

  • TTcross achieves pricing accuracy comparable to Monte Carlo while using roughly 50–100 times less walltime. performance
    Source sentence
    TTcross approach achieves pricing accuracy comparable to Monte Carlo methods while using roughly 50–100 times less walltime
  • Under high volatility (σ = 2) both TTcross and variational MPS reduce pricing error by more than an order of magnitude versus Monte Carlo. performance
    Source sentence
    under high volatility (σ = 2) both the TTcross and variational MPS methods reduce the pricing error by more than order of magnitude compared to Monte Carlo
  • The variational MPS method provides a rigorous lower bound on option prices. theoretical
    Source sentence
    the variational MPS approach not only approximates the option price efficiently but also provides a rigorous lower bound

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 35 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 1 paper in the archive (1 by independent authors): Full grid solution for multi-asset options pricing with … (uses its method or data)

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.