A Sufficient Information Approach to Decentralized Decision Making

We study a general class of decentralized dynamic decision-making problems with many agents, asymmetric information, and hidden actions. We propose the notion of sufficient information that provides a mutually consistent compression of the agents' private and common information for decision-making purposes. We define a class of strategies, called sufficient information-based (SIB) strategies, that are based on the agents' sufficient information. We show that restriction to SIB strategies is without loss of optimality in decentralized decision problems with non-strategic agents (i.e. teams). Accordingly, we provide a sequential decomposition of dynamic teams over time that specifies an algorithm for determining globally optimal strategies. For decentralized decision problems with strategic agents (i.e. games), we show that the class of SIB strategies is closed under the best response map. Consequently, we propose a notion of sufficient information-based equilibrium and provide a sequential decomposition of dynamic games over time that specifies an algorithm for determining Sufficient Information Based Perfect Bayesian Equilibria (SIB-PBE).