Conditional Linear Combination Tests for Weakly Identified Models

Section A gives empirical results for the Angrist and Krueger (1991) data and discusses implementation of PI tests and confidence sets. Section B provides further details on the derivation of the limit problems discussed in Section 2 of the paper. Section C shows that general nonlinear GMM models which are weakly identified in the sense of Stock and Wright (2000) give rise to limiting problems of the form (2). Section D concerns our linear IV simulations, gives power plots for PI tests in linear IV with homoskedastic errors, provides further information on our simulation design, and discusses our implementation of the MM1-SU, MM2-SU, QLR, and PI tests. Section E provides additional details on our implementation of PI and QLR tests in Section A. Finally, Section F discusses simulation results and derivations in a nonlinear new Keynesian Phillips curve model.