The finite horizon optimal stopping model’s various applications, mainly in American-type option pricing, required research in order to construct a solution for these problems in exact form. Dynkin’s original formulation and solution [1] uses dynamic programming methods, but rarely provides the exact analytical and numerical solution. The alternative is to approximate the model by an optimization problem with a known algorithm. Some of these techniques are obtained from iterative schemes for Markov decision processes.

In this paper, Borkar et al.’s main idea is based on the formulation of the optimal stopping problem: “The ‘primal’ formulation of this is a linear program (LP for short) over the so-called ‘occupation measures.’ Its ‘dual’ is an LP over functions” [2].

Next, the learning scheme is constructed for the linear approximation of the LP. The main part of the paper is devoted to the derivation and justification of the algorithm. The work is theoretical, with descriptions of numerical experiments related to the option pricing model.

This well-executed application of known techniques offers new, effective methods for solving optimal stopping problems. The examples presented are appealing. Although the paper lists other methods of approximation, it fails to provide a direct comparison of the proposed approach to other approaches.