Correlated equilibria in a class of repeated games with incomplete information (Q1075257)

A two-person infinitely repeated game \(\Gamma\) is modified by introducing the so-called communication device d. Before every stage in the original game \(\Gamma\), the device sends separately a signal to each player, according to some probability and some independent actions of the players transmitted to the device without informing the opponent. The modified game is denoted by \(\Gamma_ d\). The pay-offs in \(\Gamma_ d\) are defined as limits of their expected average pay-offs for the n first stages. In a usual way an associated basic probability space is constructed. The convergence problem for the expectation of the pay-offs is discussed in the appendix. Given a communication device d and the fact that player 1 is informed in \(\Gamma\) about one of the possible states of Nature in L, the set \({\mathcal V}(d)\in R^ L\times R\) of equilibrium pay- offs in \(\Gamma_ d\) is introduced. The author considers two particular devices, namely ''noisy channel'' and ''correlation device''. For the ''noisy channel'', at the first stage the action set for player 1 is L, the signal set for player 2 is J where J is a finite set of choices for player 2 in \(\Gamma\) and all other sets are singletons. A noisy channel equilibrium is an equilibrium in \(\Gamma\) in which player 1 transmits \(\ell \in L\) to the device and player 2 uses at every stage the action \(j\in J\) sent to him by the device. The set of such equilibrium points is denoted by \({\mathcal N}.\) The correlation device demands no actions of the players at its input. It sends a pair of correlated signals according to a single probability distribution. The correlation device is finite if its set of signals is finite. Two main theorems are proved. Let \({\mathcal V}=\cup_{d}{\mathcal V}(d)\) be the set of all equilibrium pay-offs associated with all conceivable extensions of \(\Gamma\). Theorem 1 states that \({\mathcal V}={\mathcal N}\) and Theorem 2 gives the existence of a finite correlation device d such that \({\mathcal V}={\mathcal V}(d).\) The paper is composed of four sections: introduction, statement of the results and their interpretation in relation to some previous results in the theory, proof of the results and finally, an interesting numerical example.