Martingale optimal transport duality (Q2664166)

The set \(\mathcal{D}([0,T];\mathbb{R}^d_+)\) of all \(\mathbb{R}^d_+\)-valued functions \(\omega:[0,T]\to\mathbb{R}^d_+\) that are continuous from the right and have finite left limits with respect to Jakubowski's \(S\)-topology is called the Skorokhod space. If \(\Omega\) is a closed subset of \(\mathcal{D}\), \(X_t(\omega)=\omega(t)\) is the coordinate map on \(\Omega\), and \(\mathbb{F}^X=(\mathcal{F}_t^X)_{t\in[0,T]}\) is the natural filtration of \(X\) given by \(\mathcal{F}^X_t=\sigma(X_s;\ s\le t)\), then \(\mathbb{F}=(\mathcal{F}_t)\) is the right-continuous filtration given by \(\mathcal{F}_t=\mathcal{F}^X_{t+}=\bigcap_{s<t}\mathcal{F}^X_s\). If \(\mathcal{B}(\Omega)\) is the set of all Borel measurable functions \(\xi:\Omega\to[-\infty,\infty]\), \(\mathcal{B}_b(\Omega)\) is the subset of bounded functions in \(\mathcal{B}(\Omega)\), \(\mathcal{B}_p(\Omega)\) is the subset of functions \(\xi\) in \(\mathcal{B}(\Omega)\) such that \(\xi(\omega)/(1+\|\omega\|^p_\infty)\) is bounded, \(\mathcal{U}_b(\Omega)\) and \(\mathcal{U}_p(\Omega)\) are the sets of all upper semicontinuous functions in \(\mathcal{B}_b(\Omega)\) and \(\mathcal{B}_p(\Omega)\), respectively, \(\mathcal{C}(\Omega)\) is the set of all real-valued continuous functions, and \(\mathcal{C}_b(\Omega)\) and \(\mathcal{C}_p(\Omega)\) are defined analogously to \(\mathcal{U}_b(\Omega)\) and \(\mathcal{U}_p(\Omega)\), then the set \(\mathcal{C}_{q,p}(\Omega)\) is defined as \[ \mathcal{C}_{q,p}(\Omega)=\{\xi\in\mathcal{C}(\Omega);\ \xi^+\in\mathcal{C}_q(\Omega) \wedge \xi^-\in\mathcal{C}_p(\Omega)\}, \] where \(\xi^+=\max(\xi,0)\) and \(\xi^-=\max(-\xi,0)\). If \(\mathcal{M}(\Omega)\) is the set of all signed Radon measures of bounded total variation on \(\Omega\) and \(\mathcal{P}(\Omega)\subset\mathcal{M}(\Omega)\) is the subset of probability measures, then, for \(Q\in\mathcal{P}(\Omega)\) and \(\xi\in\mathcal{B}(\Omega)\), the expectation \(\mathbb{E}_Q[\xi]\in[-\infty,\infty]\) is defined as \(\mathbb{E}_Q[\xi]=\mathbb{E}_Q[\xi^+]-\mathbb{E}_Q[\xi^-]\). The set of all functions \(\xi\in\mathcal{B}(\Omega)\) satisfying \(\mathbb{E}_Q[|\xi|^p]<\infty\) is denoted by \(\mathcal{L}^p(\Omega,Q)\). If \(\mathcal{G}\subset\mathcal{B}(\Omega)\) is a convex cone, then \(\mathcal{Q(G)}\) is the set of all probability measures \(Q\in\mathcal{P}(\Omega)\) such that \(\mathbb{E}_Q[\gamma]\le 0\) for all \(\gamma\in\mathcal{G}\), and the canonical map \(X\) is an \((\mathbb{F},Q)\)-martingale, that is, for every \(t\in[0,T]\), \(X_t\in\mathcal{L}^1(\Omega,Q)\) and \(\mathbb{E}_Q[Y\cdot(X_t-X_T)]=0\) for all \(\mathcal{F}_t\)-measurable \(Y\in\mathcal{B}_b(\Omega)^d\). If \(\Phi\) is convex, lower semicontinuous functional defined on \(X=\mathcal{B}_b(\Omega)\) of \(X=\mathcal{C}_b(\Omega)\), then the quotient set given by a convex cone \(\mathcal{I}\) is defined as \[ \Phi(\xi;\mathcal{I})=\inf\{c\in\mathbb{R};\ c+\ell\ge \xi\ \text{for some}\ \ell\in\mathcal{I}\},\ \xi\in X. \] In this paper, the authors study general martingale optimal transport on a subset \(\Omega\) of the Skorokhod space \(\mathcal{D}([0,T];\mathbb{R}^d_+)\). They obtain a dual representation of the Kantorovich functional which takes the form of a Choquet capacity generated by martingale measures satisfying additional constraints to ensure compatibility with the quotient sets. In the main result of the paper, the authors show that if \(\mathcal{G}\subset\mathcal{C}_{q,p}(\Omega)\) is a convex cone, and there exist \(c_q\in\mathbb{R}_+\) and \(\xi_q\in\mathcal{G}\) such that \(|X_T|^q\le c_q+\xi_q\), then \[ \begin{gathered} \Phi(\xi;\mathcal{I(G)})=\sigma_{\mathcal{Q(G)}}(\xi)\ \text{for all}\ \xi\in\mathcal{U}_p(\Omega),\ \text{and}\\ \Phi(\xi;\widehat{\mathcal{I}}(\mathcal{G}))=\sigma_{\mathcal{Q(G)}}(\xi)\ \text{for all}\ \xi\in\mathcal{B}_p(\Omega), \end{gathered} \] where \(\sigma_{\mathcal{Q(G)}}(\cdot)=\sup_{}\mathbb{E}_Q[\cdot]\) is the support functional of \(\mathcal{Q(G)}\), and \(\widehat{\mathcal{I}}(\mathcal{G})\) is the Fatou closure of \(\mathcal{I}(\mathcal{G})\).