Total: 1
This paper proposes to project and expand the conditional mean function of the structural error given the regressors in an endogenous regression under consideration. As the projection process is semiparametric, we define this procedure as a semiparametric projection (SP) method to address endogeneity in regression models by internally constructed instrumental variables. The SP method is applicable to many classes of regression models associated with endogeneity, such as linear, nonlinear, and non- and semi-parametric models, and provides a simple and computationally tractable alternative to conventional instrumental variable approaches available from the existing literature. This paper establishes identification conditions and derives the asymptotic properties of the resulting estimators. It then proposes a simple LASSO selection method to examine the finite-sample performance of both the proposed method and the established theory by simulated and real data examples.