Dynamic Choice of Information Sources

Abstract

An agent has to choose among a finite set of actions with payoffs depending on an unknown state of the world with two (possibly correlated) binary components, (θ1, θ2), θi∈{0,1}. Before making the choice, the agent can learn about the state by dynamically allocating his limited attention over two information sources, modelled as search processes for conclusive evidence of θ1=1 or θ2=1. In the special case where θ1+θ2 ≤1, the agent searches for the most probable evidence up to some time t*, and after that he continues by challenging his current beliefs in the most effective way. The agent never changes his allocation of attention after t*. When the state (1,1) is possible, the attention rule before t* is different. Sufficient conditions for t*>0 are given. A negative correlation between the state components increases t*.

Keywords: dynamic information acquisition, Poisson process, limited attention, correlation, media consumption

JEL Classification: D83