Alethograph

Notes on Backward Stochastic Differential Equations for Computing XVA

5 min read

Summary

This paper provides a rigorous BSDE-based framework for understanding XVA (X-valuation adjustment) pricing of OTC derivatives in a post-crisis setting. The pre-crisis paradigm priced derivatives at the risk-neutral expectation under a single risk-free rate; the post-crisis reality requires accounting for counterparty credit risk, funding costs, and collateral rates, producing correction terms collectively called XVA = CVA - DVA + FVA + ColVA + … The authors set up the pricing/hedging problem for a defaultable derivative on a progressively enlarged filtration generated by Brownian motion and two default indicator processes (investor default at tau_1, counterparty default at tau_2). The derivative’s maturity is the random horizon tau_1 ^ tau_2 ^ T, and the hedger’s self-financing portfolio value satisfies a BSDE with random horizon driven by (W, M_1, M_2). Through a reduction technique, this BSDE on the enlarged filtration G is mapped to a standard BSDE on the reference Brownian filtration F, which in the Markovian case connects to a semilinear PDE via the nonlinear Feynman-Kac formula.

The hedging problem is formulated as finding replicating portfolios for both seller and buyer, yielding two BSDEs with nonlinear drivers f^+ and f^- that differ by terms involving the funding and repo lending-borrowing spreads epsilon_f and epsilon_r. The paper establishes an explicit sufficient condition (Theorem 4) for the arbitrage-free property: any price in the interval [Y^-(0), Y^+(0)] is arbitrage-free. Treating the spreads as small parameters, the authors show (Theorem 5) that the zeroth-order approximations Y^{0,pm}, which solve linear BSDEs, are themselves arbitrage-free and admit a closed-form decomposition (Proposition 3) into the base price V plus individual XVA decomposition into valuation adjustment terms — DVA, CVA, FVA, and ColVA. A first-order perturbation expansion (Proposition 4) improves the approximation to O(epsilon^2), paving the way for numerical implementation via deep BSDE method solvers.

Key Contributions

  • Generalises the XVA framework of Bichuch, Capponi, and Sturm (2018) to multiple risky assets and stochastic volatility/interest rate/hazard rate models
  • Derives seller and buyer BSDEs (37) and (39) with nonlinear drivers on a progressively enlarged filtration, with explicit reduction to the reference filtration
  • Provides an explicit sufficient condition (Theorem 4) for the arbitrage-free property of the bilateral hedging price interval [Y^-(0), Y^+(0)]
  • Interprets practitioner XVA formula as a zeroth-order approximation of the theoretical fair price, with a precise O(epsilon) error bound (Theorem 5)
  • Derives closed-form decomposition of the zeroth-order price into V + VA_1 + VA_2 + VA_3 + VA_4 + VA_5, recovering DVA, CVA, FVA, and ColVA as special cases
  • Computes first-order perturbed BSDEs (Proposition 4) giving O(epsilon^2) accuracy

Methodology

The paper works on a probability space carrying an n-dimensional Brownian motion W and two independent exponential random variables E_1, E_2. Default times tau_i are constructed as the first passage times of integrated hazard rate processes. The progressively enlarged filtration G = F v H incorporates both Brownian and default information. The key technical device is the reduction theorem (Theorem 2): any BSDE on (Omega, F, P, G) with random horizon tau_1 ^ tau_2 ^ T reduces to a standard BSDE on (Omega, F, P, F), with the jump components U_i determined by phi_i(t) - Y_bar(t). In the Markovian setting, the nonlinear Feynman-Kac formula (Theorem 3) connects the reduced BSDE to a semilinear PDE. The comparison theorem for BSDEs is used repeatedly: to order Y^- ⇐ Y^+ (establishing the bid-ask spread), to prove monotonicity of the price in the payoff (yielding the arbitrage-free result), and to sandwich Y^{0,pm} between Y^- and Y^+. The perturbation analysis treats epsilon_f, epsilon_r as small parameters and expands the BSDE solution in powers of epsilon.

Key Findings

  • The arbitrage-free condition requires hazard rates to dominate certain combinations of funding-repo-risk-free rate spreads (condition 48), plus rf^+ >= rcol^- (condition 49); violating these is realistic, suggesting that admitting limited arbitrage may be necessary in practice (Remark 14)
  • When all rates coincide (rD = rf = rr = rcol), the model collapses to the classical Black-Scholes-Merton setting with a single risk-free rate
  • The zeroth-order approximation Y^{0,pm} solves a linear BSDE, hence admits a closed-form representation under a Girsanov change of measure
  • The XVA decomposition V + DVA + CVA + FVA + ColVA emerges as a special case of the general formula when rr^0 = rf^0 = rD
  • Positive homogeneity of the BSDE drivers ensures price scales linearly with notional (Remark 15)
  • The first-order correction (Proposition 4) is itself a linear BSDE with a source term depending on the zeroth-order solution, making it amenable to Monte Carlo or deep learning computation

Important References

  1. Bichuch, Capponi, Sturm (2018) “Arbitrage-free XVA” — the primary predecessor model generalised here (50 citations)
  2. El Karoui, Peng, Quenez (2000) “Backward stochastic differential equations in finance” — foundational BSDE reference for the nonlinear Feynman-Kac formula (2618 citations)
  3. Pardoux and Peng (1990) “Adapted solution of backward stochastic equation” — introduced general nonlinear BSDEs (2903 citations)
  4. Pham (2010) “Stochastic control under progressive enlargement of filtrations” — source for the filtration enlargement and reduction techniques (66 citations)
  5. Bielecki and Rutkowski (2004) “Credit Risk: Modeling, Valuation, and Hedging” — comprehensive credit risk reference used throughout (1194 citations)
  6. Crépey and Song (2016) “Counterparty risk and funding: immersion and beyond” — related BSDE reduction results (49 citations)
  7. Gregory (2015) “The XVA Challenge” — practitioner-oriented comprehensive XVA reference (84 citations)

Atomic Notes

paper

Graph View

Backlinks