A Specification Test for Nonparametric Instrumental Variable Regression

We consider testing for correct specification of a nonparametric instrumental variable regression. First we study the notion of correct specification, misspecification and overidentification in this ill-posed inverse problem setting. Second we study a test statistic based on the empirical minimum di...

Full description

Saved in:
Bibliographic Details
Published inAnnals of economics and statistics no. 128; pp. 151 - 202
Main Authors Patrick Gagliardini, Olivier Scaillet
Format Journal Article
LanguageEnglish
Published GENES 01.12.2017
Subjects
Instrumental variables
Instrumental variables estimation
Inverse problems
Null hypothesis
Online AccessGet full text

Cover

More Information
Summary:We consider testing for correct specification of a nonparametric instrumental variable regression. First we study the notion of correct specification, misspecification and overidentification in this ill-posed inverse problem setting. Second we study a test statistic based on the empirical minimum distance criterion corresponding to the conditional moment restriction evaluated with a Tikhonov Regularized estimator of the functional parameter. The test statistic admits an asymptotic normal distribution under the null hypothesis, and the test is consistent under global alternatives. A bootstrap procedure is available to get simulation based critical values. Finally, we explore the finite sample behavior with Monte Carlo experiments, and provide an empirical illustration for an estimated Engel curve. JEL: C13, C14, C15, D12 / KEY WORDS: Specification Test, Nonparametric Regression, Instrumental Variables, Minimum Distance, Tikhonov Regularization, Ill-posed Inverse Problems, Generalized Method of Moments, Bootstrap, Engel Curve.
ISSN:2115-4430
1968-3863
DOI:10.15609/annaeconstat2009.128.0151