An optimal transport approach for the multiple quantile hedging problem

Basic Information

• Paper ID: 2308.01121
• Title: An optimal transport approach for the multiple quantile hedging problem
• Authors: Cyril Bénézet, Jean-François Chassagneux, Mohan Yang
• Classification: math.PR (Probability Theory), q-fin.CP (Computational Finance)
• Submission Date: August 2, 2023 (arXiv submission)
• Paper Link: https://arxiv.org/abs/2308.01121

Abstract

This paper investigates the Multiple Quantile Hedging (MQH) problem, a class of partial hedging problems that encompasses quantile hedging (Föllmer & Leukert 1999) and profit-and-loss (P&L) matching problems (Bouchard & Vu 2012) as special cases. In complete nonlinear markets, the authors demonstrate that this problem can be reformulated as a Monge optimal transport problem. Building on this observation, a Kantorovich version of the problem is introduced and the equivalence of the two formulations is established. In the linear case, the MQH problem can be viewed as a semi-discrete optimal transport problem, and a dual formulation is derived. The absence of a duality gap is proven, enabling the design of numerical methods based on stochastic gradient algorithms for computing MQH prices.

Research Background and Motivation

Problem Context

1. Limitations of superhedging: Traditional superhedging strategies, while providing complete risk immunity, typically require substantial initial capital, resulting in prohibitively high premiums that undermine the seller's competitiveness.
2. Need for partial hedging: Partial hedging strategies reduce initial premiums by accepting some residual risk exposure. However, potential negative P&L outcomes necessitate systematic risk control methodologies.
3. Inadequacies of existing approaches:
   • Quantile hedging problems are primarily solved via dynamic programming and PDE methods, but numerical implementation is challenging
   • P&L matching problems lack a unified theoretical framework
   • Effective methods for handling multiple quantile constraints are lacking

Research Motivation

• Provide a unified framework for addressing diverse partial hedging problems
• Leverage the powerful tools of optimal transport theory for analysis and solution
• Develop practical numerical algorithms, particularly those capable of handling path-dependent options

Core Contributions

1. Definition of a new problem class: Introduces the Multiple Quantile Hedging (MQH) problem, unifying and generalizing quantile hedging and P&L matching problems.
2. Establishment of optimal transport connection: First demonstration of equivalence between the MQH problem and Monge-type optimal transport problems under nonlinear expectations.
3. Comprehensive theoretical analysis:
   • Proves equivalence of Monge and Kantorovich problem values
   • Establishes duality theory without duality gap in linear markets
4. Practical numerical methods: Designs stable and efficient numerical solution schemes based on stochastic gradient ascent algorithms.

Methodology Details

Problem Definition

Given a probability measure $\mu \in P(\llbracket N \rrbracket)$ and a monotone random vector $G = (G^n)_{1 \leq n \leq N}$, the multiple quantile hedging price is defined as:

$V^{MQH}(G,\mu) := \inf H^{MQH}(G,\mu)$

where

$H^{MQH}(G,\mu) := \{y \in \mathbb{R} \mid \exists \nu \in H^2(\mathbb{F},P;\mathbb{R}^m), \forall n \in \llbracket N \rrbracket, P(Y_T^{y,\nu} \geq G^n) \geq \bar{F}_\mu(n)\}$

Here $Y^{y,\nu}$ is the wealth process driven by initial capital $y$ and strategy $\nu$.

Model Architecture

1. Financial Market Model

• Risk-free asset: $dS_t^0 = r_t S_t^0 dt$, $S_0^0 = 1$
• Risky assets: $dS_t = \text{diag}(S_t)(\beta_t dt + \sigma_t dW_t)$
• Wealth dynamics: $Y_t = y - \int_0^t f(s, Y_s, \nu_s)ds + \int_0^t \nu_s^T dW_s$

where $f$ encodes the market's nonlinear characteristics.

2. Relaxed Monge Problem

Define the relaxed Monge problem: $V^{RM}(G,\mu) = \inf_{\chi \in T_\mu^+(\mathcal{F}_T)} Y_0[G^\chi]$

where $T_\mu^+(\mathcal{F}_T)$ is the set of $\mathcal{F}_T$-measurable random variables with distributional stochastic dominance $\mu$.

3. Kantorovich Problem

The corresponding Kantorovich problem is: $V^{KP}(G,\mu) = \inf_{P \in P_\mu(\mathcal{F}_T)} Y_0\left[\sum_{n=1}^N G^n P^n\right]$

Technical Innovations

1. Optimal transport perspective: First connection between partial hedging and optimal transport theory, providing a novel analytical framework.
2. Transport under nonlinear expectations: Extension of classical optimal transport theory to nonlinear expectation settings, where target distributions satisfy stochastic dominance constraints rather than being fixed.
3. Equivalence proofs: Constructive proofs establishing $V^{MQH} = V^{RM} = V^{KP}$, providing theoretical foundation for numerical solution.

Experimental Setup

Numerical Methods

In the linear market case, utilizing the dual representation: $V^{MQH}(G,\mu) = \sup_{\Phi \in (\mathbb{R}_+)^N} \left\{E\left[\min_{1 \leq n \leq N}(H^n - \Phi^n)\right] + \sum_{n=1}^N \Phi^n p^n\right\}$

where $H^n = \Gamma_T G^n$, and $\Gamma_T$ is the discount factor in the linear case.

Algorithm Implementation

Stochastic gradient ascent algorithm implemented using ADAM optimizer:

• Batch size: 64-256
• Learning rate: 0.01
• Convergence criterion: $|\zeta_m - \zeta_{m-1}| < 10^{-6}$

Test Cases

1. Quantile hedging: Quantile hedging of call/put options
2. P&L distribution hedging: Shape control of P&L distributions
3. Multiple quantile hedging: Multi-constraint hedging of call spread options

Experimental Results

Main Results

1. Quantile Hedging Verification

• Perfect alignment with theoretical values from Föllmer-Leukert formula
• High precision maintained at extreme quantile values ($p$ near 0 or 1)
• Stable algorithm convergence with standard deviation controlled in range 0.001-0.01

2. P&L Distribution Hedging

Comparison between SG-solver and OT-solver (semi-analytical solution based on optimal transport):

Results demonstrate high consistency between methods, validating SG-solver accuracy.

3. Computational Efficiency

• Superior numerical stability compared to PDE methods
• Natural extension to non-Markovian settings and path-dependent derivatives
• Capability to handle arbitrary finite numbers of quantile constraints

Case Study

Call Spread Option

Market parameters: $S_0 = 100$, $r = 0$, $\sigma = 0.2$, $\beta = 0.1$, $K = 100$

MQH prices under different quantile constraints:

• Replication constraint $G_1$: 3.67
• Replication constraint $G_2$: 6.97
• Replication constraint $G_3$: 7.97
• Multi-quantile combinations demonstrate price monotonicity and constraint effectiveness

Related Work

Main Research Directions

1. Quantile hedging: Pioneering work by Föllmer & Leukert (1999), subsequently developed through stochastic target problems and BSDE methods
2. P&L matching: Distribution control approach proposed by Bouchard & Vu (2012)
3. Optimal transport applications in finance: Primarily focused on robust pricing and model uncertainty problems

Advantages of This Work

• Provides a unified theoretical framework
• Introduces novel mathematical tools (nonlinear optimal transport)
• Develops practical numerical algorithms
• Extends problem applicability

Conclusions and Discussion

Main Conclusions

1. The MQH problem can be completely characterized and solved through optimal transport theory
2. Absence of duality gap in linear markets provides theoretical guarantees for numerical computation
3. Stochastic gradient-based algorithms demonstrate excellent performance in practical applications

Limitations

1. Complete market assumption: Theoretical analysis restricted to complete markets, whereas real markets are typically incomplete
2. Computational complexity: Computational complexity may increase significantly with the number of constraints
3. Nonlinear case: Numerical algorithms for nonlinear markets require further development

Future Directions

1. Incomplete markets: Extension to incomplete market settings, better reflecting MQH's potential as an alternative pricing principle
2. Parameter uncertainty: Consideration of parameter uncertainty under the physical measure P
3. Time-dependent constraints: Investigation of time-dependent multiple quantile hedging problems

In-Depth Evaluation

Strengths

1. Theoretical innovation: First establishment of deep connection between partial hedging and optimal transport theory, providing novel perspective for the field
2. Mathematical rigor: Complete and rigorous proofs, with constructive proofs of equivalence theorems being particularly valuable
3. Strong practicality: Provided numerical algorithms are stable and efficient, capable of handling complex practical problems
4. Unified framework: Unifies seemingly unrelated problems under a single theoretical framework

Weaknesses

1. Application scope: Complete market assumption limits practical applicability
2. Complexity analysis: Lacks theoretical analysis of algorithm complexity
3. Limited testing scale: Numerical experiments are relatively limited in scale, lacking validation on large-scale real data

Impact

1. Academic value: Opens new directions for cross-disciplinary research between financial mathematics and optimal transport theory
2. Practical applications: Provides new tools for financial institutions' risk management
3. Methodological contribution: Demonstrates enormous potential of optimal transport in financial problems

Applicable Scenarios

• Partial hedging strategy design for derivatives
• Portfolio risk budget management
• Quantile risk control for insurance products
• Risk-constrained optimization in quantitative trading

References

Key references for this paper include:

1. Föllmer, H., & Leukert, P. (1999). Quantile hedging. Finance and Stochastics, 3(3), 251-273.
2. Bouchard, B., & Vu, T. N. (2012). A stochastic target approach for P&L matching problems. Mathematics of Operations Research, 37(3), 526-558.
3. El Karoui, N., Peng, S., & Quenez, M. C. (1997). Backward stochastic differential equations in finance. Mathematical finance, 7(1), 1-71.
4. Villani, C. (2021). Topics in optimal transportation. American Mathematical Society.

This paper makes significant contributions at both theoretical and applied levels. The innovation of introducing optimal transport theory to partial hedging problems is particularly noteworthy. While limitations exist regarding complete market assumptions, the work establishes a solid foundation for future development in this field.