On a randomized strategy in Neveu's stopping problem (Q1081199): Difference between revisions

\textit{E. B. Dynkin} [Dokl. Akad. Nauk SSSR 185, 16-19 (1969; Zbl 0186.253)] presented the following problem in game theory: two players observe a stochastic sequence X(n), \(n=1,2,... \). Each of them chooses a stopping time; let \(\lambda\) (resp. \(\mu)\) be the stopping time chosen by the first (resp. second) player. The payoff is then: \[ R(\lambda,\mu)= \begin{cases} X(\lambda)& \text{if \(\lambda\leq \mu,\)} \\ X(\mu)& \text{if \(\lambda >\mu\).} \end{cases} \] Player 1 seeks to maximize the expected payoff, and player 2 seeks to minimize it. \textit{J. Neveu} [Discrete parameter martingales. (1975; Zbl 0345.60026)] modified this problem as follows: there are two random sequences X(n) and Y(n), with X(n)\(\leq Y(n)\) for each n, and the payoff equals: \[ R(\lambda,\mu)= \begin{cases} X(\lambda)& \text{if \(\lambda\leq \mu,\)} \\ Y(\mu)& \text{if \(\lambda >\mu.\)} \end{cases} \] For this paper, three random sequences X(n), Y(n) and W(n) are considered, and the payoff equals: \[ R(\lambda,\mu)= \begin{cases} X(\lambda)& \text{if \(\lambda <\mu,\)} \\ W(\lambda)& \text{if \(\lambda =\mu,\)} \\ Y(\mu)& \text{if \(\lambda >\mu.\)} \end{cases} \] The set of strategies (stopping times) is extended to include randomized strategies. First, a finite horizon problem is considered: under the condition \(X(n)\leq W(n)\leq Y(n)\), the existence of the value of the game and of an optimal pure strategy is proved; then, for the infinite horizon problem with a discount factor, a condition is given for the existence of a game value.