Estimating Choice Models with Unobserved Expectations over Attributes

Agents often make choices by forming expectations about attributes, but such expectations are usually unobserved by researchers. We develop two methods for estimating discrete choice models where agents use unobserved heterogeneous information sets to form expectations. Preferences are point-identified using a finite mixture approximation of the unobserved information structure or set-identified with partial information. Both methods apply to individual- and market-level data without imposing strong assumptions on how expectations are formed. We revisit two empirical applications that confirm the importance of accounting for unobserved information: firms' revenue expectations when exporting and consumers' fuel cost expectations when purchasing cars.