2025-11-17T18:52:13.444121

Martingale Optimal Transport and Martingale Schrödinger Bridges for Calibration of Stochastic Volatility Models

Zitridis
Motivated by recent developments in the calibration of stochastic volatility models (SVMs for short), we study continuous-time formulations of martingale optimal transport and martingale Schrödinger bridge problems. We establish duality formulas and also provide alternative proofs, via different techniques, of duality results previously established in the mathematical finance literature. Applications include calibration of SVMs to SPX options, as well as joint calibration to both SPX and VIX options.
academic

Martingale Optimal Transport and Martingale Schrödinger Bridges for Calibration of Stochastic Volatility Models

Basic Information

  • Paper ID: 2510.10860
  • Title: Martingale Optimal Transport and Martingale Schrödinger Bridges for Calibration of Stochastic Volatility Models
  • Author: Antonios Zitridis
  • Classification: math.OC (Optimization and Control), math.PR (Probability)
  • Publication Date: October 12, 2025 (arXiv preprint)
  • Paper Link: https://arxiv.org/abs/2510.10860

Abstract

Motivated by recent developments in the field of stochastic volatility models (SVMs) calibration, this paper investigates the continuous-time formulation of martingale optimal transport and martingale Schrödinger bridge problems. The paper establishes dual formulas and provides alternative proofs for previously established dual results in the mathematical finance literature through different techniques. Applications include calibrating SVMs to SPX options, as well as simultaneous calibration to both SPX and VIX options.

Research Background and Motivation

Core Problem

The core problem addressed in this paper is the calibration of stochastic volatility models. Since the introduction of the Black-Scholes model, the academic community has been devoted to developing sophisticated volatility models that accurately capture market dynamics. To correct systematic errors and ensure accurate predictions, these models must be appropriately calibrated to vanilla options (such as S&P 500 options and VIX options).

Problem Significance

  1. Practical Application Needs: The accuracy of option pricing in financial markets directly impacts trading strategies and risk management
  2. Theoretical Refinement: Arbitrage-free model calibration methods are needed to ensure consistency of financial theory
  3. Computational Efficiency: Existing methods face challenges in numerical implementation, requiring better dual formulas to support numerical approximation

Limitations of Existing Methods

  1. The connection between continuous-time martingale optimal transport problems and their discrete versions remains unclear
  2. Lack of a unified theoretical framework for handling simultaneous calibration of SPX and VIX options
  3. Existing proof techniques primarily rely on the Fenchel-Rockafellar theorem, lacking alternative approaches

Research Motivation

The author employs techniques from mean field game theory, particularly the Von-Neumann minimax theorem, to provide new theoretical analysis methods for martingale optimal transport and martingale Schrödinger bridge problems.

Core Contributions

  1. Established dual formulas for continuous-time martingale optimal transport problems with a new proof method based on the Von-Neumann minimax theorem
  2. Provided new dual formula proofs for martingale Schrödinger bridge problems using technical approaches different from existing literature
  3. Resolved the joint SPX-VIX calibration problem, providing rigorous dual formulas for martingale Schrödinger bridges with VIX constraints
  4. Applied viscosity solution theory to handle Hamilton-Jacobi equations with Dirac delta terms
  5. Provided explicit expressions for optimal solutions, including Radon-Nikodym derivatives of optimal probability measures

Methodology Details

Problem Formulation

Martingale Optimal Transport Problem

For given probability measures μ0,μ1,μ2∈P(R)\mu_0, \mu_1, \mu_2 \in P(\mathbb{R}) satisfying convex order relations μ0≤cμ1≤cμ2\mu_0 \leq_c \mu_1 \leq_c \mu_2, time parameters 0<T1<T20 < T_1 < T_2 and T0∈[0,T2]T_0 \in [0, T_2], the martingale optimal transport problem is defined as:

V(T0,μ0,μ1,μ2):=inf⁡P∈Pm{EP[∫T0T2L(σt2)dt]}V(T_0, \mu_0, \mu_1, \mu_2) := \inf_{P \in P_m} \left\{ E^P \left[ \int_{T_0}^{T_2} L(\sigma_t^2) dt \right] \right\}

where PmP_m is the set of probability measures satisfying martingale conditions and marginal distribution constraints.

Martingale Schrödinger Bridge Problem

Given a reference measure P0P_0 and constraint set Ct0(μ1,μ2)C_{t_0}(\mu_1, \mu_2), the martingale Schrödinger bridge problem is:

DP0:=inf⁡P∈Ct0(μ1,μ2)H(P∣P0)D_{P_0} := \inf_{P \in C_{t_0}(\mu_1, \mu_2)} H(P|P_0)

where H(P∣P0)H(P|P_0) is the relative entropy.

Model Architecture

Stochastic Volatility Model

Consider the continuous-time stochastic volatility model:

dX_t = \sigma(X_t, Y_t) dW_t \\ dY_t = b(X_t, Y_t) dt + \tau_1(X_t, Y_t) dW_t + \tau_2(X_t, Y_t) dW_t^{\perp} \end{cases}$$ where $W, W^{\perp}$ are independent Brownian motions and $\sigma(x,y) = x\tilde{\sigma}(y)$. #### Hamilton-Jacobi Equation The solution to the dual problem is characterized through the following Hamilton-Jacobi equation: $$\begin{cases} -\partial_t u + H\left(\frac{D^2u}{2}\right) = \delta_{T_1}(t)u_1 \\ u(T_2, x) = u_2(x) \end{cases}$$ where $H(a) = \sup_{b \geq 0}\{-ab - L(b)\}$ is the Hamiltonian. ### Technical Innovations #### 1. Application of Von-Neumann Minimax Theorem Unlike the traditional Fenchel-Rockafellar approach, this paper employs the Von-Neumann minimax theorem to establish duality relations, which requires verification of: - Concavity-convexity of functionals - Compactness conditions - Lower semicontinuity #### 2. Viscosity Solution Theory for Dirac Delta Terms The paper develops viscosity solution theory for handling Hamilton-Jacobi equations with Dirac delta functions, defining jump conditions: $$u(T_1^-, x) - u(T_1^+, x) = u_1(x)$$ #### 3. Relaxation of VIX Constraints For the joint calibration problem, the original VIX equality constraint: $$L^P(V_P) = \mu_3$$ is relaxed to a convex order constraint: $$L^P(V_P) \leq_{c,l} \mu_3$$ This makes the problem more tractable mathematically. ## Experimental Setup ### Theoretical Verification Framework This paper is primarily theoretical research, with experimental setup focused on theorem proofs and mathematical verification: #### Assumptions - **(A1)**: $L: \mathbb{R} \to \mathbb{R}$ is a convex function and $p$-coercive - **(A2)**: Hamiltonian $H$ is a $C^1$ function - **(A3)**: Lipschitz and growth conditions on coefficient functions - **(A4)**: Additional regularity conditions for uniqueness #### Proof Strategy 1. Establish feasibility using the mimicking theorem 2. Apply Von-Neumann theorem to exchange inf and sup 3. Utilize stochastic control theory to characterize value functions 4. Handle jump conditions through viscosity solution theory ## Main Results ### Theorem 1.2 (Martingale Optimal Transport Dual Formula) Under assumptions (A1) and (A2), for convex-ordered probability measures $\mu_0 \leq_c \mu_1 \leq_c \mu_2$, the dual formula holds: $$V(T_0, \mu_0, \mu_1, \mu_2) = \sup_{u_1, u_2 \in \text{Lip}} \left\{ \int u(T_0, x) d\mu_0(x) - \int u_2(x) d\mu_2(x) - \mathbf{1}_{[0,T_1]}(T_0) \int u_1(x) d\mu_1(x) \right\}$$ where $u$ is the viscosity solution to the Hamilton-Jacobi equation. ### Theorem 1.4 (Martingale Schrödinger Bridge Dual Formula) Under assumption (A3): $$\inf_{P \in C_{t_0}(\mu_1, \mu_2)} H(P|P_0) = \sup_{u_1, u_2 \in \text{Lip}} \left\{ -\mathbf{1}_{[0,T_1]}(t_0) \int u_1 d\mu_1 - \int u_2 d\mu_2 + u(t_0, X_{t_0}, Y_{t_0}) \right\}$$ with explicit expression for the optimal measure: $$\frac{dP^*}{dP_0}\bigg|_{F_{T_2}} = e^{-\int_{t_0}^{T_2} \tau_2(X_t, Y_t) \partial_y u^*(t, X_t, Y_t) dW_t^{\perp} - \frac{1}{2} \int_{t_0}^{T_2} \tau_2^2(X_t, Y_t) |\partial_y u^*(t, X_t, Y_t)|^2 dt}$$ ### Theorem 1.5 (Joint SPX-VIX Calibration) For the relaxed VIX constraint, the dual formula is established: $$D'_{P_0} = \sup_{u_1, u_2, u_3} \left\{ -\int u_1 d\mu_1 - \int u_2 d\mu_2 - \int u_3 d\mu_3 + u(t_0, X_{t_0}, Y_{t_0}) \right\}$$ where $u_3$ is a convex lower-bounded function and $u$ is characterized through an optimal control problem. ## Theoretical Analysis ### Existence and Uniqueness - Feasibility is established through Strassen's theorem and Brenier's theorem - Existence and uniqueness of solutions to Hamilton-Jacobi equations are guaranteed using standard viscosity solution theory results - For cases with state constraints, results from [DLL06] are employed ### Regularity Results **Corollary C.5**: Under appropriate assumptions, the viscosity solution $u(t, \cdot)$ is uniformly Lipschitz continuous in time. ### Continuity Results **Lemma B.3**: When $\tau_2$ is constant, the function $\Phi(x,y)$ is continuous, which is crucial for establishing regularity of the value function. ## Related Work ### Optimal Transport Theory - **Classical Theory**: Monge-Kantorovich problem and its Benamou-Brenier continuous-time formulation - **Martingale Constraints**: Semimartingale optimal transport by Mikami-Thieullen and Tan-Touzi - **Continuous Time**: Continuous-time martingale optimal transport by Henry-Labordère et al. ### Schrödinger Bridge Theory - **Classical Schrödinger Bridges**: Minimizing relative entropy between two probability measures - **Martingale Constraint Extensions**: Martingale Schrödinger bridges introduced by Labordère - **Financial Applications**: Applications to VIX calibration by Guyon et al. ### Numerical Methods Although this paper focuses on theory, the dual formulas provide foundation for numerical methods: - GLW17, GLOW22: Numerical simulation based on dual formulas - Guy22: Computational implementation of joint SPX-VIX calibration ## Conclusions and Discussion ### Main Conclusions 1. **Unified Theoretical Framework**: Provides unified duality theory for martingale optimal transport and martingale Schrödinger bridges 2. **New Proof Techniques**: Von-Neumann theorem provides a more direct proof path than the Fenchel-Rockafellar theorem 3. **Practical Applications**: Resolves important joint calibration problems in finance ### Limitations 1. **VIX Constraint Relaxation**: Only handles convex order constraints rather than equality constraints 2. **Regularity Assumptions**: Requires relatively strong coefficient regularity assumptions 3. **Numerical Implementation**: Does not provide specific numerical algorithms and efficiency analysis ### Future Directions 1. **Complete VIX Constraints**: Investigate methods for handling original equality constraints 2. **Numerical Algorithms**: Develop efficient numerical methods based on dual formulas 3. **Model Extensions**: Generalize to more general stochastic volatility models ## In-Depth Evaluation ### Strengths 1. **Theoretical Innovation**: Introduces Von-Neumann theorem to provide new perspectives on classical problems 2. **Mathematical Rigor**: Complete proofs and detailed technical details 3. **Practical Value**: Addresses actual calibration problems in finance 4. **Clear Presentation**: Good mathematical exposition and logical structure ### Weaknesses 1. **Application Scope**: Relaxation of VIX constraints limits practical applications 2. **Computational Complexity**: Does not analyze algorithm computational complexity 3. **Empirical Validation**: Lacks verification with real market data ### Impact 1. **Theoretical Contribution**: Provides new proof techniques for optimal transport theory 2. **Practical Value**: Significant implications for model calibration in financial engineering 3. **Methodology**: Von-Neumann approach may be applicable to other related problems ### Applicable Scenarios - Option pricing model calibration in financial institutions - Volatility modeling in risk management - Theoretical analysis tools in academic research ## Technical Details Supplement ### Viscosity Solution Definition **Definition C.1**: A function $u$ is a viscosity solution to equation (1.10) if it satisfies the corresponding Hamilton-Jacobi equations on intervals $(T_1, T_2]$ and $[0, T_1)$ respectively, and satisfies the jump condition at $T_1$. ### Key Lemmas **Lemma 2.4**: Weak convergence properties of probability densities, crucial for establishing conditions of Von-Neumann theorem. **Propositions 2.2, 2.3**: Compactness results for Fokker-Planck equation solutions, ensuring convergence of sequences in optimization problems. --- This paper makes important contributions at the intersection of theoretical mathematics and financial applications, providing new mathematical tools for stochastic volatility model calibration. While certain technical details have limitations, its theoretical value and potential application prospects merit attention.