Please use this identifier to cite or link to this item: https://hdl.handle.net/10419/97395 
Year of Publication: 
2013
Series/Report no.: 
cemmap working paper No. CWP48/13
Publisher: 
Centre for Microdata Methods and Practice (cemmap), London
Abstract: 
This paper asks which aspects of a structural Nonparametric Instrumental Variables Regression (NPIVR) can be identified well and which ones cannot. It contributes to answering this question by characterizing the identified set of linear continuous functionals of the NPIVR under norm constraints. Each element of the identified set of NPIVR can be written as the sum of a common 'identifiable component' and an idiosyncratic 'unidentifiable component'. The identified set for any continuous linear functional is shown to be a closed interval, whose midpoint is the functional applied to the 'identifiable component'. The formula for the length of the identified set extends the popular omitted variables formula of classical linear regression. Some examples illustrate the wide applicability and utility of our identification result, including bounds and a new identification condition for point-evaluation functionals. The main ideas are illustrated with an empirical application of the effect of children on labor-market outcomes for women.
Subjects: 
Instrumental Variables
Nonparametric
Semiparametric
Bounds
Linear functional
JEL: 
C26
C14
C21
Persistent Identifier of the first edition: 
Document Type: 
Working Paper

Files in This Item:
File
Size
584.93 kB





Items in EconStor are protected by copyright, with all rights reserved, unless otherwise indicated.