Robust Model Misspecification and Paradigm Shifts

This paper studies the forms of model misspecification that are more likely to persist when an agent compares her subjective with competing models. The learns about action-dependent outcome distribution and makes decisions repeatedly. Aware potential misspecification, she uses a threshold rule switch between models according how well they fit data. A is globally robust if it can against every locally nearby under priors. main result provides simple characterization based on set BerkNash equilibria induce. I then use these results provide first learning foundations for persistence systemic biases in two canonical applications.