This paper considers delta estimators of a Radon-Nicodym derivative of a probability function with respect to a measure. Sufficient conditions for asymptotic unbiasedness and global rates of convergence, which can be improved by imposing differentiability conditions on the estimated curves, are provided. A bias reduction technique is proposed, and the application of the results to regression estimation is discussed. The sufficient conditions for asymptotic unbiasedness are checked for some broad classes of nonparametric estimators.
展开▼
