Identified Regions and Inference in Panel Data Roy Models

Joint with: Elie Tamer

This paper explores the identifiabilty of regression coefficients in a panel data Roy model. Our approach is based on deriving the form of conditional moment inequal-

ities which can be used to infer the identified region of the parameter space.

Our set of moment inequalities is complete in the sense the bounds we attain the parameters are sharp. A method based on a transformation to what we call unconditional

integrated moment inequalities is proposed, to which sample analogs can be used to consistently estimate the identified region. Sufficient conditions for point identification

and asymptotic properties of the proposed estimation procedure are derived.

An extension to identification in cross sectional Roy models is considered, and an inference method based on conditional moment inequalities after pairwise differencing is proposed. Finite sample properties of the proposed methods are explore through a small scale simulation study and an empirical illustration.

For more information, contact Frank Schorfheide.