Double robust inference for continuous updating GMM

TL;DR

This paper introduces the double robust Lagrange multiplier (DRLM) test for hypothesis testing in GMM models that remains valid under misspecification and weak identification, enhancing robustness in empirical analysis.

Contribution

It proposes the DRLM statistic, which provides robust inference for the pseudo-true parameters in GMM, addressing issues of misspecification and weak identification.

Findings

DRLM test has a chi-squared limiting distribution robust to misspecification.

Application to return on education shows effectiveness under weak identification.

Analysis of risk premia demonstrates robustness in empirical settings.

Abstract

We propose the double robust Lagrange multiplier (DRLM) statistic for testing hypotheses specified on the pseudo-true value of the structural parameters in the generalized method of moments. The pseudo-true value is defined as the minimizer of the population continuous updating objective function and equals the true value of the structural parameter in the absence of misspecification.\nocite{hhy96} The (bounding) chi-squared limiting distribution of the DRLM statistic is robust to both misspecification and weak identification of the structural parameters, hence its name. To emphasize its importance for applied work, we use the DRLM test to analyze the return on education, which is often perceived to be weakly identified, using data from Card (1995) where misspecification occurs in case of treatment heterogeneity; and to analyze the risk premia associated with risk factors proposed in Adrian et al. (2014) and He et al. (2017), where both misspecification and weak identification need to be addressed.