When conducting regression analysis, econometricians often face the situation where some relevant regressors are unavailable in the data set at hand. This paper shows how to construct a new class of nonparametric proxies by combining the original data set with one containing the missing regressors. Imputation of the missing values is done using a nonstandard kernel adapted to mixed data. We derive the asymptotic distribution of the resulting semiparametric two-sample estimator of the parameters of interest and show, using Monte Carlo simulations, that it dominates the solutions involving instrumental variables and other parametric alternatives. An application to the PSID and NLS data illustrates the importance of our estimation approach for empirical research.
Research related links:
Link to work
GAUSS-Codes for Simulations and Applications