WP 21-30 – We present a geometric approach to the finite Rational Inattention (RI) model, recasting it as a convex optimization problem with reduced dimensionality that is well-suited to numerical methods.
This working paper and Working Paper 21-29 both supersede Working Paper 20-24 – Rational Inattention via Ignorance Equivalence.
We provide an algorithm that outperforms existing RI computation techniques in terms of both speed and accuracy. We also introduce methods to quantify the impact of numerical inaccuracy on the behavioral predictions and to produce robust predictions regarding the most frequently implemented actions.
