We consider the value of persistent information in strictly competitive situations, formalized as stochastic zero-sum games where only the maximizer ob-serves the state that evolves according to an ergodic Markov operator. We say that operator Q is better for the maximizer than operator P if the value of the game under Q is higher than under P regardless of the stage game. We show that this defines a partial order on the space of ergodic Markov operators, and provide a full characterization of this partial order. An i.i.d. state is the best case for the informed player; however, a perfectly persistent state is not necessarily the worst case. The analysis relies on a novel characterization of the value of a stochastic game with incomplete information. Our results can alternatively be interpreted as pertaining to the limit of the minmax value in repeated Bayesian games with Markov types. 1.

We study repeated Bayesian games with communication and observable actions in which the players' privately known payoffs evolve according to an irreducible Markov chain whose transitions are independent across players. Our main result implies that, generically, any Pareto-efficient payoff vector above a stationary minmax value can be approximated arbitrarily closely in a perfect Bayesian equilibrium as the discount factor goes to 1. As an intermediate step, we construct an approximately efficient dynamic mechanism for long finite horizons without assuming transferable utility.

We consider learning and signalling in a dynamic Cournot oligopoly where firms have private information about their production costs and only observe the market price, which is subject to unobservable demand shocks. An equilibrium is Markov if play depends on the history only through the firms’ beliefs about costs and calendar time. We characterize symmetric linear Markov equilibria as solutions to a boundary value problem. In every such equilibrium, given a long enough horizon, play converges to the static complete information outcome for the realized costs, but each firm only learns its competitors’ average cost. The weights assigned to costs and beliefs under the equilibrium strategies are non-monotone over time. We explain this by decomposing incentives into signalling and learning, and discuss implications for prices, quantities, and welfare.

The copyright to this Article is held by the Econometric Society. It may be downloaded, printed and reproduced only for educational or research purposes, including use in course packs. No downloading or copying may be done for any commercial purpose without the explicit permission of the Econometric Society. For such commercial purposes contact the Office of the Econometric Society (contact information may be found at the website