Published online by Cambridge University Press: 13 September 2011
This paper derives the limiting distributions ofalternative jackknife instrumental variables (JIV)estimators and gives formulas for accompanyingconsistent standard errors in the presence ofheteroskedasticity and many instruments. Theasymptotic framework includes the many instrumentsequence of Bekker (1994, Econometrica 62,657–681) and the many weak instrument sequence ofChao and Swanson (2005, Econometrica 73,1673–1691). We show that JIV estimators areasymptotically normal and that standard errors areconsistent provided that asn→∞, whereKn andrn denote,respectively, the number of instruments and theconcentration parameter. This is in contrast to theasymptotic behavior of such classical instrumentalvariables estimators as limited information maximumlikelihood, bias-corrected two-stage least squares,and two-stage least squares, all of which areinconsistent in the presence of heteroskedasticity,unlessKn/rn→0.We also show that the rate of convergence and theform of the asymptotic covariance matrix of the JIVestimators will in general depend on the strength ofthe instruments as measured by the relative ordersof magnitude of rn andKn.
Earlier versions of this paper were presented atthe NSF/NBER conference on weak and/or manyinstruments at MIT in 2003 and at the 2004 wintermeetings of the Econometric Society in San Diego,where conference participants provided many usefulcomments and suggestions. Particular thanks areowed to D. Ackerberg, D. Andrews, J. Angrist, M.Caner, M. Carrasco, P. Guggenberger, J. Hahn, G.Imbens, R. Klein, N. Lott, M. Moriera, G.D.A.Phillips, P.C.B. Phillips, J. Stock, J. Wright,two anonymous referees, and a co-editor forhelpful comments and suggestions.