Information and strategies in dynamic games
Abstract
We extend to the setting of stochastic dynamic games a theorem of Kuhn, and use it to prove the existence of a saddle point in a suitable class of strategies. We then particularize this result to the situation where one of the players has full information to show existence of a saddle point in another class of strategies exhibiting a constant dimension sufficient statistic. A dynamic programming-like algorithm is naturally associated with this class of strategies, and was proposed in a previous paper in a sufficient condition setting. For this same class of games, we give an example of another use of the main theorem, leading to a different dynamic programming-like algorithm.