Bayesian information topologies (Q922258)

A state of the world \(\omega\) is drawn to some known prior distribution, \(\mu\). Agents receive their information about \(\omega\) and form their posterior distributions, \(P_{\omega}\), by conditioning on the sub- \(\sigma\)-fields that represent their information. The mapping \(\omega \mapsto P_{\omega}\) contains all of the economically relevant aspects of an agent's information. `Bayesian' topologies on these mappings are defined by saying that information structures are close if there is only small probability that the posterior distributions are far apart. This approach leads to a characterization of the relation between partition representations of information and \(\sigma\)-field representations of information.