RECURSIVE ALGORITHMS FOR OPTIMAL SOLUTIONS OF STOCHASTIC TEAMS WITH DECENTRALIZED INFORMATION.

The authors consider a class of stochastic team and nonzero-sum game problems with more than two agents who have access to decentralized information and may build their own subjective probability models to be used in the decision process. There is, in general, no compatibility between different models built by different agents, and this makes the available theory on teams and games inapplicable to this problem. The authors also discuss different equilibrium solutions to the team and game problems in this multimodeling framework, and develop convergent algorithms which would lead to such an equilibrium under a number of conditions and for different probabilistic models. As a byproduct of this analysis, a recursive algorithm is obtained which provides a solution to quadratic teams when the underlying distributions are not Gaussian.