Robust best choice problem (Q2216194)

The full information best choice (FIBC) problem is one of the best known optimal stopping problems in discrete time. The agent is interested in detecting the maximum of a finite sequence of independent, uniformly distributed random variables, thus identifying the stopping time that maximizes the probability measure. In this paper a robust multiple priors version of the classical FIBC problem is formulated and solved. The only thing the agent knows about the distribution that drives the process is that it belongs to a set of priors. The random variables are not independent nor identically distributed under every measure in the set of priors and it is demonstrated that the problem cannot be reduced to a single prior optimal stopping problem in which the process consists of independent random variables. Similarly as in the classical case, the solution (optimal stopping time) can be fully characterized by a set of decreasing thresholds. The author proposes a novel and quite general construction of a set of priors that satisfies time consistency, thus solving a very general version of the robust best choice problem. The construction represents a notable contribution to the theory of optimal stopping under multiple priors. The history dependent minimizing measure is characterized and sensitivity analysis on two examples is performed. The answers for questions ``What is the effect of introducing multiple priors to the FIBC problem?'' and ``How does the optimal stopping time change once ambiguity is introduced?'' are given.