RESEARCH PAPER NO. 862

ASSESSING INSTRUMENTAL VARIABLE RELEVANCE: AN ALTERNATIVE MEASURE AND SOME EXACT FINITE SAMPLE THEORY

BY

D. S. POSKITT & C. L. SKEELS

NOVEMBER 2002

Department of Economics. University of Melbourne. Melbourne Victoria 3010 Australia.

Existing measures for the relevance of instrumental variables tend to focus on the ability of the instrument set to predict a single endogenous regressor, even if there is more than one endogenous regressor in the equation of interest.  We propose new measures of instrument relevance in the presence of multiple endogenous regressors, taking both univariate and multivariate perspectives, and develop the accompanying exact finite sample distribution theory in each case.  In passing, the paper also explores relationships that exist between the measures proposed here and other statistics that have been proposed elsewhere in the literature.  These explorations highlight the close connection between notions of instrument relevance, identification and specification testing in simultaneous equations models.