When an economic agent makes a choice, stochastic models predicting those choices can be updated. The structural assumptions embedded in the prior model condition the updated one, to the extent that the same evidence produces different predictions even when previous ones were identical. We provide a general framework for models of stochastic choice allowing for arbitrary forms of (structural) updating and show that different models can be sharply separated by their structural properties, leading to axiomatic characterizations. Our framework encompasses Bayesian updating given beliefs over deterministic preferences (as implied by popular random utility models) and standard neuroeconomic models of choice, which update decision values in the brain through reinforcement learning.
Stochastic preferences Bayesian learning logit choice reinforcement neuroeconomic theory