Belief Disagreement, Instability, and Uncertainty under Stochastic Recall

Expectations are typically modeled as functions of an agent’s information set. Using preregistered evidence, we show that beliefs depend instead on the information agents retrieve when forming them. Even after observing identical histories of inflation or stock returns, individuals recall different observations, and their beliefs track what they retrieve. Selective recall therefore generates disagreement, within-person belief instability, and heterogeneity in subjective uncertainty without any new information. We formalize these patterns in a parsimonious model in which agents store observed signals, retrieve each only probabilistically, and update optimally using recalled rather than objectively available information. The model yields sharp predictions for how beliefs vary with the composition of stored information, recall probabilities, and contextual cues. Large-scale experiments that hold information fixed and intervene only after exposure confirm these predictions, while survey evidence on macroeconomic and financial expectations supports their external relevance. Disagreement, instability, and uncertainty are thus structural features of belief formation rather than mere measurement error.