Paper: arXiv 2401.13314

Authors: Lokman Abbas-Turki, Stéphane Crépey, Botao Li, Bouazza Saadeddine

Abstract

Motivated by the equations of cross valuation adjustments (XVAs) in the realistic case where capital is deemed fungible as a source of funding for variation margin, we introduce a simulation/regression scheme for a class of anticipated BSDEs, where the coefficient entails a conditional expected shortfall of the martingale part of the solution. The scheme is explicit in time and uses neural network least-squares and quantile regressions for the embedded conditional expectations and expected shortfall computations. An a posteriori Monte Carlo validation procedure allows assessing the regression error of the scheme at each time step. The superiority of this scheme with respect to Picard iterations is illustrated in a high-dimensional and hybrid market/default risks XVA use-case.

Complexity vs Empirical Score

  • Math Complexity: 8.5/10
  • Empirical Rigor: 6.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper introduces an explicit scheme for complex anticipated BSDEs with heavy mathematical derivations and advanced concepts, demonstrating high mathematical complexity. It provides a simulation/regression scheme with a posteriori Monte Carlo validation and an empirical benchmark in a 36-dimensional XVA use-case, indicating substantial empirical rigor.

Research Flowchart

  flowchart TD
  Goal["Research Goal<br>Pathwise XVA via Fungible Capital"]
  Input["Data Inputs<br>Market Models & Counterparty Data"]
  
  Method["Methodology<br>Explicit Scheme via Neural Networks"]
  Process["Computational Process<br>Monte Carlo + Regression Steps"]
  
  Residual["A Posteriori Validation<br>Regression Error Assessment"]
  Outcome["Findings<br>Superiority over Picard Iterations"]
  
  Goal --> Input
  Input --> Method
  Method --> Process
  Process --> Residual
  Process --> Outcome

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 an explicit simulation/regression scheme for anticipated BSDEs arising in XVA computations, using neural network least‑squares and quantile regressions.

Source sentence
we introduce a simulation/regression scheme for a class of anticipated BSDEs, where the coefficient entails a conditional expected shortfall of the martingale part of the solution. The scheme is explicit in time and uses neural network least-squares and quantile regressions for the embedded conditional expectations and

Assumes 5

  • Coefficients b(t,x) and σ(t,x) are Lipschitz in x uniformly in t and have linear growth in x. model
    Source sentence
    for coefficients b(t, x) and σ(t, x) Lipschitz in x uniformly in t ∈[0, T] and with linear growth in x.
  • Every (F,P) martingale in S^2 starting from 0 admits a representation Z·W + U·µ. model
    Source sentence
    starting from 0 has a representation of the form
  • The map Φ_h is Lipschitz in its arguments. model
    Source sentence
    Φ_h is assumed Λ_h^Φ Lipschitz in the sense mimicking (9) that … ≤Λ_h^Φ …
  • The process γ(b)(t,X_t)(P_c J(t,X_t)M_t^c(t,X_t)) is in H^2 (e.g., γ(b) bounded). model
    Source sentence
    Assumption 2.2(ii) holds provided the process γ(b)(t, X_t)(P_c J(t
Show 1 more
  • The XVA equations have zero terminal conditions. model
    Source sentence
    as all our XVA equations are endowed with zero terminal conditions

Data

Synthetic data only.

Costs

Not applicable The paper reports no trading or execution result.

Finds 4

  • The explicit scheme dominates the implicit/Picard scheme in computation time, being roughly 3–4 times faster. performance
    Source sentence
    the Picard scheme comes at the cost of j = 3-4 times longer computations (for reference, the explicit scheme takes 7mins 48secs using the finest time grid, i.e. θ = 8, on an NVidia A100 GPU).
  • Both schemes produce mutually validating XVA numbers, with the Picard iteration converging after a few steps to results visually close to the explicit scheme. theoretical
    Source sentence
    toward a solution visually very close, already for j = 4, to the one provided by the explicit scheme
  • Mean and quantiles of CVA, FVA, KVA and EC learned by the explicit scheme are reported across different time‑step sizes. statistical
    Source sentence
    Figure 1: Mean and quantiles of CVA, FVA, KVA and EC learned by the explicit scheme at t = T/2 for different sizes of the time step.
  • Local regression errors (L2) decrease with finer time discretization, confirming the accuracy of the regression components. statistical
    Source sentence
    L2 error against conditional expectation at time t

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 26 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 3 papers in the archive (2 by independent authors)

Code

https://github.com/BouazzaSE/NeuralXVA — 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.