Similarity of Information in Games

with Joyee Deb and Aditya Kuvalekar

under review

Supported by National Science Foundation award 2417694

[Link to paper] 

Abstract: What is a reasonable notion for comparing similarity of information across agents in any Bayesian game? At the very least, more similar information across agents should aid coordination. However, it turns out that existing stochastic orders that compare the interdependence of joint distributions do not have this intuitive property. We propose a new class of orders called “Concentration Along Diagonal” (CAD) that compares information similarity in Bayesian games. When information becomes more CAD-similar, each agent believes it is more likely that others have also received the same information. We show that CAD-similarity is equivalent to aiding coordination in canonical binary-action coordination games. That is, more CAD-similar information aids coordination in all such games, and if an information change aids coordination in all such games, then it must be more CAD-similar. We apply CAD in other well-known games such as congestion, collective action, or auctions.

Presentation (by one of the authors): Bonn University, Brown University, Carnegie Mellon University, Cornell University, Indian Statistical Institute Delhi, National University of Singapore, Ohio State University, Oxford University, Paris School of Economics, Royal Holloway University, Toulouse School of Economics, University of Rochester, University of Surrey,