We consider the estimation of a semiparametric regression model where data is independently and identically distributed. Our primary interest is on the estimation of the parameter vector, where the associated regressors are correlated with the errors and contain both continuous and discrete variables. We propose three estimators by adapting Robinson's (1988) and Li and Stengos' (1996) framework and establish their asymptotic properties. They are asymptotically normally distributed and correctly centered at the true value of the parameter vector. Among a class of semiparametric IV estimators with conditional moment restriction, the first two are efficient under conditional homoskedasticity and the last one is efficient under heteroskedasticity. They allow the reduced form to be nonparametric, are asymptotically equivalent to semiparametric IV estimators that optimally select the instrument and reach the semiparametric efficiency bounds in Chamberlain (1992). A Monte Carlo study is performed to shed light on the finite sample properties of these competing estimators. Its applicability is illustrated with an empirical data set.
