We study the description and value of information in zero-sum games. We define a series of informational relations between information schemes, and show that informational equivalence classes are captured by canonical information structures. Moreover, two information schemes induce the same value in every game if and only if they are informationally equivalent. We prove the existence of a revealing game in which unique optimal strategies are homeomorphic to canonical types.
We study the robustness of equilibria with regards to small payoff perturbations of the dynamic game. We find that complete penal codes, that specify players’ strategies after every history, have only limited robustness. For some generic games, no complete codes exist that are robust to even arbitrarily small perturbations. We define incomplete penal codes as partial descriptions of equilibrium strategies and introduce a notion of robustness for incomplete penal codes. We prove a Folk Theorem in robust incomplete codes that generates a Folk Theorem in a class of stochastic games.
We show how group testing can be used in three applications to multiply the efficiency of tests: estimation of virus prevalence, releasing group to the work force, and testing for individual infectious status. For an infection level around 2%, group testing could potentially allow to save 94% of tests in the first application, 95% in the second, and 85% in the third one.
We show how group testing can be used in three applications to multiply the efficiency of tests against COVID-19: estimating virus prevalence, releasing group to the work force, and testing for individual infectious status. For an infection level around 2%, group testing could potentially allow to save 94% of tests in the first application, 95% in the second, and 85% in the third one.
We study the impact of manipulating the attention of a decision-maker who learns sequentially about a number of items before making a choice. Under natural assumptions on the decision-maker’s strategy, forcing attention toward one item increases the likelihood of its being chosen.
We consider an agent who acquires information on a state of nature from an information structure before facing a decision problem. How much information is worth depends jointly on the decision problem and on the information structure. We represent the decision problem by the set of possible payoffs indexed by states of nature. We establish and exploit the duality between this set on one hand and the value of information function, which maps beliefs to expected payoffs under optimal actions at these beliefs, on the other. We then derive global estimates of the value of information of any information structure from local properties of the value function and of the set of optimal actions taken at the prior belief only.