Stopping with expectation constraints: 3 points suffice (Q2316590)

The paper deals with a problem of optimally stopping a process with a stopping time satisfying an expectation constraint. Let \((Y_t)_{t\in\Re^+}\) be a one-dimensional regular continuous strong Markov process with respect to a right-continuous filtration \((\mathcal{F}_t)\). The state space \(J\subset\Re\) is assumed to be an open, half-open or closed interval. The payoff \(f : \Re\rightarrow\Re\) is a Borel-measurable having regularity conditions. Let us denote by \(\mathcal{T} (\mathbf{T})\) the set of \((\mathcal{F}_t)\)-stopping times such that \(\mathbf{E}[\tau]\leq \mathbf{T}\in \Re^+\). The problem considered in the paper has the form: \(\max_{\tau\in \mathcal{T} (\mathbf{T})} \textbf{E}[f(Y_\tau)]\). It is shown that to obtain a solution of such constrained optimization problem it suffices to consider only stopping times such that the law of the process at the stopping time is a weighted sum of 3 Dirac measures. The idea for proving a reduction to 3 Dirac measures is based on rewriting the considered constrained stopping problem as a linear optimization problem over a set of probability measures. Recent results of \textit{S. Ankirchner} et al. [Bernoulli 21, No. 2, 1067--1088 (2015; Zbl 1328.60101)] and \textit{D. Hobson} [Electron. J. Probab. 20, Paper No. 83, 26 p. (2015; Zbl 1328.60104)] on the Skorokhod embedding problem characterizing the set \(\mathcal{A}(\mathbf{T})\) of probability distributions that can be embedded into \(Y\) with stopping times having expectation smaller than or equal to \(\mathbf{T}\). As for standard linear problems the maximal value of the optimization is attained by extreme points. The extreme points of \(\mathcal{A}(\mathbf{T})\) turn out to be contained in the set of probability measures that can be written as weighted sums of at most 3 Dirac measures.