Bootstrap inference for misspecified moment condition models (Q1753973)

The authors carry out a detailed theoretical study of several analytic and bootstrap test procedures under potentially misspecified moment condition models. It turns out that none of the existing bootstrap methods performs better than the corresponding analytic (asymptotic) test in all cases. Based on this theoretical analysis, the authors propose to use several bootstrap methods in combination with each other. This strategy leads to improved accuracy under correct specification as well as consistency under misspecification. Finally, the theoretical findings are illustrated by means of computer simulations and an analysis of real data.