A0354
Title: Two-sample estimation of varying coefficient models via nearest neighbor matching
Authors: Masayuki Hirukawa - Setsunan University (Japan) [presenting]
Artem Prokhorov - University of Sydney (Australia)
Abstract: Economists often face the situation in which all necessary variables must be collected from more than one source when running a regression. We investigate the problem of estimating varying coefficient models using the combined sample that is constructed from two samples via the nearest neighbor matching. Our particular focus is on the kernel-smoothed local linear estimation of the varying coefficient. It is demonstrated that the local linear estimator using matched samples is inconsistent, as is the case with the ordinary least squares estimator of the linear regression model using matched samples. If only a few variables are used to impute the missing data, then it is possible to correct for the bias. We propose a bias-correction method for the local linear estimation and explore asymptotic properties of the bias-corrected estimator. Monte Carlo simulations confirm that the bias correction works very well in such cases.
