Core Concepts
The authors provide a fundamental theorem of asset pricing and a superhedging duality in a setting that combines dependence uncertainty with additional information on the dependence structure in the form of known prices for multi-asset options. They solve the resulting optimization problem using a penalization approach combined with a deep learning approximation.
Abstract
The article presents a hybrid approach between model-uncertainty and model-free/data-driven methods in finance. The authors consider the classical setting of dependence uncertainty, where the marginals are known but the dependence structure is unknown. They extend this setting by incorporating additional information in the form of traded multi-asset options.
The key contributions are:
Derivation of a fundamental theorem of asset pricing and a superhedging duality in this setting.
Characterization of the optimal measures and trading strategies.
Numerical method based on a penalization approach and deep learning approximation to compute the model-free bounds.
Numerical experiments using artificial data to evaluate the impact of additional information and the scalability of the method.
The numerical results show that the method is fast and accurate, and the computational time scales linearly with the number of assets. The authors also observe that "relevant" information, i.e. prices of derivatives with the same payoff structure as the target payoff, are more useful than other information and should be prioritized.
Stats
The initial values, variances and correlation matrix used in the numerical experiments are:
S0 = [10, 10, 10]
σ = [0.3, 0.4, 0.5]
ρ = [[1, 0.5, 0.5], [0.5, 1, 0.5], [0.5, 0.5, 1]]