Abstract
We present some Monte Carlo results on three semi parametric models, namely, partly linear, error-in-variables, and generalised least squares models. The results favour the use of computational expensive cross-validation criterion for bandwidth selection only when the relative sample size is large.