Rational Inattention to Discrete Choices: A New Foundation for the Multinomial Logit Model

We apply the rational inattention approach to information frictions to a discrete choice problem. The rationally inattentive agent chooses how to process information about the unknown values of the available options to maximize the expected value of the chosen option less an information cost. We solve the model analytically and find that if the agent views the options as equivalent a priori, then the agent chooses probabilistically according to the multinomial logit model, which is widely used to study discrete choices. When the options are not symmetric a priori, the agent incorporates prior knowledge of the options into the choice in a simple fashion that can be interpreted as a multinomial logit in which an option's a priori attractiveness shifts its perceived value. Unlike the multinomial logit, this model predicts that duplicate options are treated as a single option.