Title: How much should attention cost?

Abstract: Using the motivation from rational inattention (Sims, 1998, 2003), I consider a model of dynamic information acquisition based on sequential marginal experimentation. Taking the limit as individual experiments become completely uninformative, there is a unique stochastic differential equation representing the information process, yielding a particular posterior-separable measure of uncertainty that can be used to easily calculate costs. One can then use this cost function to predict choices in decision problems. I discuss extensions to heterogeneous marginal costs and multiple states.