Stability of the weak martingale optimal transport problem (Q6139683)

Let \(X\) and \(Y\) be Polish spaces endowed with the compatible and complete metrics \(d_{X}\) and \(d_{Y}\), \(\mu \in \mathcal{P}(X)\) the set of probability measures on \(X\), \(\nu \in \mathcal{P}(Y)\) and \(C:X\times \mathcal{P} (Y)\rightarrow \mathbb{R}_{+}\) be a nonnegative measurable function which is convex with respect to its second argument. \(\Pi (\mu ,\nu )\) denotes the set of couplings between \(\mu \) and \(\nu \), that is, \(\pi \in \Pi (\mu ,\nu ) \) if and only if \(\pi \in \mathcal{P}(X\times Y)\) is such that for any measurable subsets \(A\subset X\) and \(B\subset Y\), \(\pi (A\times Y)=\mu (A)\) and \(\pi (X\times B)=\nu (B)\). The weak optimal transport problem consists in the minimization problem: \(V_{C}(\mu ,\nu )=\inf_{\pi \in \Pi (\mu ,\nu )}\int_{X}C(x,\pi _{x})\mu (dx)\), where, for all \(\pi \in \Pi (\mu ,\nu )\), \( (\pi _{x})_{x\in \mathbb{R}}\) denotes a disintegration of \(\pi \) with respect to its first marginal: \(\pi (dx,dy)=\mu (dx)\pi _{x}(dy)\). The authors also consider the martingale optimal transport problem: \(\inf\pi \in \Pi _{M}(\mu ,\nu )\int_{\mathbb{R}\times \mathbb{R}}c(x,y)\pi (dx,dy)\), where \(\mu ,\nu \in \mathcal{P}^{1}(\mathbb{R})=\{p\in \mathcal{P} (X):\int_{X}d_{X}(x,x_{0})p(dx)<\infty \}\), \(c:\mathbb{R}\times \mathbb{ R\rightarrow R}_{+}\) is a measurable cost function, and \(M(\mu ,\nu )=\{\pi =\mu \times \pi _{x}\in \Pi (\mu ,\nu ):\int_{\mathbb{R}}y\pi _{x}(dy)=x\) \( \mu \)-a.s.\(\}\). They introduce a weak version of this problem with cost functions \(C\) of the form \(C:\mathbb{R}\times \mathcal{P}(\mathbb{R} )\rightarrow \mathbb{R}_{+}\), and which consists in the minimization problem: \(V_{C}^{M}(\mu ,\nu )=\inf\pi \in \Pi _{M}(\mu ,\nu )\int_{\mathbb{R} \times \mathbb{R}}C(x,\pi _{x})\mu (dx)\). The first main result of the paper proves a stability result for the weak martingale optimal transport problem, assuming that \(C:\mathbb{R}\times \mathcal{P}^{1}(\mathbb{R})\rightarrow \mathbb{R}\) is continuous, convex with respect to the second argument and there exists a constant \(K>0\) such that for \((x,p)\in \mathbb{R}\times \mathcal{P}^{1}(\mathbb{R})\): \(\left\vert C(x,p)\right\vert \leq K(\left\vert x\right\vert +\int_{Y}\left\vert y\right\vert p(dy)\). For \(k\in \mathbb{N}\), let \(\mu ^{k},\nu ^{k}\in \mathcal{P}^{1}(\mathbb{R})\), (resp. satisfying \(\mu ^{k}\leq _{c}\nu ^{k}\)) converge in \(\mathcal{P}^{1}(\mathbb{ R})\) to \(\mu \) and \(\nu \), respectively. Then \(\lim_{k\rightarrow \infty }V_{C}(\mu ^{k},\nu ^{k})=V_{C}(\mu ,\nu )\) (resp. \(\lim_{k\rightarrow \infty }V_{C}^{M}(\mu ^{k},\nu ^{k})=V_{C}^{M}(\mu ,\nu )\)). There exist minimizers \(\pi ^{k,\ast }\in \Pi (\mu ^{k},\nu ^{k})\) (resp. \(\pi ^{k,\ast }\in \Pi _{M}(\mu ^{k},\nu ^{k})\)) of \(V_{C}(\mu ^{k},\nu ^{k})\) (resp.\(V_{C}^{M}(\mu ^{k},\nu ^{k})\)) and any accumulation point of \((\pi ^{k,\ast })_{k\in \mathbb{N}}\) for the weak convergence topology is a minimizer of \(V_{C}(\mu ,\nu )\) (resp. \(V_{C}^{M}(\mu ,\nu )\)). More generally, the authors introduce the set \(\mathcal{P}^{r}(\mathbb{R})\) of all probability measures on \(X\) with finite \(r\)th moment: \(\mathcal{P}^{r}(\mathbb{R})=\{p\in \mathcal{P}(X):\int_{X}d_{X}^{r}(x,x_{0})p(dx)<\infty \}\), equipped with the weak topology induced by \(\Phi ^{r}(X)=\{f\in C(X):\exists \alpha >0,\forall x\in X,\left\vert f(x)\right\vert \leq \alpha (1+d_{X}^{r}(x,x_{0}))\}\). A compatible metric is given by the \(r\)-Wasserstein distance \(\mathcal{W}_{r}\) , that the authors adapt to the present case for \(\pi ,\pi ^{\prime }\in \mathcal{P}^{r}(X)\) as: \(\mathcal{AW}_{r}(\pi ,\pi ^{\prime })=\inf_{\chi \in \Pi (\mu ,\mu ^{\prime })}\int_{X\times X}(d_{X}^{r}(x,x^{\prime })+\mathcal{ W}_{r}^{r}(\pi _{x},\pi _{x^{\prime }}^{\prime }))\chi (dx,dx^{\prime }))^{1/r}\). The authors then extend the spaces to the case where \( f:X\rightarrow \lbrack 1,+\infty )\) is continuous as: \(\mathcal{P} _{f}(X)=\{p\in \mathcal{P}(X):p(f)<+\infty \}\), that they equip with an appropriate convergence for sequences. They prove stability results for the weak optimal transport problem and the weak martingale optimal transport problem. They deduce the stability of the superreplication bound for VIX futures and the stability of the stretched Brownian motion, and they derive a monotonicity principle for the weak martingale optimal transport problem.