Semiparametric estimation with missing covariates

This paper considers estimation in semiparametric models when some of the covariates are missing at random. The paper proposes an iterative estimator based on inverse probability weighting and local linear estimation of the nonparametric component. The resulting estimator is very general and can be used in the context of semiparametric maximum likelihood, quasi likelihood and robust estimation. The paper establishes the asymptotic normality of the estimator using both nonparametric and parametric estimation of the unknown probability weights. Two general examples illustrate the theory and Monte Carlo simulations show that the proposed estimator has good finite sample properties.