Trinity tests of functions for conditional moment models (Q2181719)

Consider a conditional moment model which includes both finite-dimensional parameters and (infinite-dimensional) functions as parameters. This paper begins by establishing conditions under which a Gaussian approximation of functionals of sieve generalized method-of-moments estimators holds. The author goes on to analyse three test statistics for hypothesis testing in this setting: one based on weighted quadratic forms, a second based on a uniform quasi-likelihood ratio statistic, and a third based on a uniform Lagrange multiplier statistic. These test statistics are shown to be asymptotically equivalent, and their bootstrap consistency is also established. These results are interpreted in the context of construction of confidence intervals and uniform confidence bands for functionals of the unknown parameters.