A CONSISTENT SPECIFICATION TEST FOR MODELS DEFINED BY CONDITIONAL MOMENT RESTRICTIONS

This article addresses statistical inference in models defined by conditional moment restrictions. Our motivation comes from two observations. First, generalized method of moments, which is the most popular methodology for statistical inference for these models, provides a unified methodology for statistical inference, but it yields inconsistent statistical procedures. Second, consistent specification testing for these models has abandoned a unified approach by regarding as unrelated parameter estimation and model checking. In this article, we provide a consistent specification test, which allows us to propose a simple unified methodology that yields consistent statistical procedures. Although the test enjoys optimality properties, the asymptotic distribution of the considered test statistic depends on the specific data generating process. Therefore, standard asymptotic inference procedures are not feasible. Nevertheless, we show that a simple original wild bootstrap procedure properly estimates the asymptotic null distribution of the test statistic.