Paper: arXiv 2511.20837

Authors: Nicolas Baradel

Abstract

In incomplete financial markets, pricing and hedging European options lack a unique no-arbitrage solution due to unhedgeable risks. This paper introduces a constrained deep learning approach to determine option prices and hedging strategies that minimize the Profit and Loss (P&L) distribution around zero. We employ a single neural network to represent the option price function, with its gradient serving as the hedging strategy, optimized via a loss function enforcing the self-financing portfolio condition. A key challenge arises from the non-smooth nature of option payoffs (e.g., vanilla calls are non-differentiable at-the-money, while digital options are discontinuous), which conflicts with the inherent smoothness of standard neural networks. To address this, we compare unconstrained networks against constrained architectures that explicitly embed the terminal payoff condition, drawing inspiration from PDE-solving techniques. Our framework assumes two tradable assets: the underlying and a liquid call option capturing volatility dynamics. Numerical experiments evaluate the method on simple options with varying non-smoothness, the exotic Equinox option, and scenarios with market jumps for robustness. Results demonstrate superior P&L distributions, highlighting the efficacy of constrained networks in handling realistic payoffs. This work advances machine learning applications in quantitative finance by integrating boundary constraints, offering a practical tool for pricing and hedging in incomplete markets.

Complexity vs Empirical Score

  • Math Complexity: 8.0/10
  • Empirical Rigor: 4.0/10
  • Quadrant: Lab Rats — high math complexity, limited empirical testing

Why this score: The paper is mathematically dense, featuring stochastic calculus, PDE theory, and constrained neural network architectures for boundary value problems. However, the empirical component is limited to conceptual numerical experiments on synthetic data (e.g., Equinox option, jumps) without code, live market data, or transaction cost details, making it less backtest-ready.

Research Flowchart

  flowchart TD
  A["Research Goal:<br>Pricing & Hedging European Options<br>in Incomplete Markets"] --> B["Methodology:<br>Constrained Deep Learning<br>Single Network for Price & Hedge"]
  B --> C["Input Data:<br>Asset Paths<br>Option Payoffs<br>Market Parameters"]
  C --> D["Computational Process:<br>Minimize P&L Variance<br>Enforce Self-Financing Constraint<br>Embed Payoff Structure"]
  D --> E["Key Outcomes:<br>Superior P&L Distribution<br>Effective Handling of Non-Smooth Payoffs<br>Robust Hedging Strategies"]

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 introduces a constrained deep learning approach that jointly learns option prices and hedging strategies by embedding the terminal payoff condition.

Source sentence
This paper introduces a constrained deep learning approach to determine option prices and hedging strategies that minimize the Profit and Loss (P&L) distribution around zero.

Assumes 5

  • The framework assumes two tradable assets: the underlying and a liquid call option capturing volatility dynamics. market
    Source sentence
    Our framework assumes two tradable assets: the underlying and a liquid call option capturing volatility dynamics.
  • The market dynamics are driven by a Brownian motion. model
    Source sentence
    We consider a general incomplete market driven by Brownian motion.
  • The hedging portfolio must be self‑financing, starting from the initial premium. execution
    Source sentence
    The hedging portfolio is required to be self-financing, starts from the initial premium, and aims to
  • Training data are generated via independent Monte Carlo simulation of the tradable assets on a discrete time grid. data
    Source sentence
    We generate n independent Monte Carlo paths of the tradable assets Z on a time grid 0 = t0 < t1 < · · · < tm = T, together with fixed or randomly drawn contract parameters...
Show 1 more
  • Option payoffs may be non‑smooth (e.g., vanilla calls, digital options) and are imposed as a terminal condition on the network. other
    Source sentence
    the (possibly non‑smooth) terminal condition p(T, z) = g(z).

Data

  • Monte Carlo simulated paths of tradable assets synthetic unknown
    Source sentence
    We generate n independent Monte Carlo paths of the tradable assets Z on a time grid 0 = t0 < t1 < · · · < tm = T, together with fixed or randomly drawn contract parameters...

Costs

Unclear No cost treatment was found in the passages the extractor read; that is not evidence that costs are not modeled.

Finds 4

  • Embedding the terminal condition in a constrained network yields superior P&L distributions compared to unconstrained networks. performance
    Source sentence
    Results demonstrate superior P&L distributions, highlighting the efficacy of constrained networks in handling realistic payoffs.
  • For simple options the method reduces the standard deviation of the P&L by up to 20%. performance
    Source sentence
    For simple options, this approach significantly improved hedging performance, even for smooth payoffs, reducing the standard deviation of the P&L by up to 20%.
  • The single‑network approach outperforms the two‑network strategy, achieving a P&L standard deviation up to 10% lower when the terminal condition is embedded. performance
    Source sentence
    The single network approach proved more accurate, achieving a P&L standard deviation up to 10% lower when embedding the terminal condition.
  • Robustness tests with volatility jumps show the Zero‑Target method remains effective under moderate jump intensity, though performance degrades under severe misspecification. performance
    Source sentence
    Robustness tests under a model with jumps further demonstrated the resilience of the Zero Target method, particularly when trained with moderate jump intensity, although performance declined under severe model misspecifications.

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.

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.