Consistent Inference in Models Defined by COnditional Moment Restrictions: an Alternative to GMM

This article introduces a unified methodology for estimating and testing nonlinear econometric models defined by conditional moment restrictions. These models are very common in econometrics, such as nonlinear rational expectation models. The current approach for inference in these models is the generalized method of moments (GMM) methodology, as proposed by Hansen and Singleton (1982). Although GMM provides a unified methodology for statistical inference that is simple to implement, it may yield inconsistent statistical procedures because it just employs a finite number of moments. This is a very important theoretical and applied problem, as illustrated by a simplified consumption-based asset pricing model. Contrary to GMM, the methodology proposed in this article delivers consistent statistical procedures because it employs an infinite number of moments that characterizes the conditional moment. In addition, the proposed methodology is widely applicable for general time series data and easy to implement. In particular, the proposed specification test relies on a novel and very simple wild bootstrap procedure.