A shape constrained smoother: simulation study (Q1965981)

The paper deals with the estimation of a functional parameter in a simple nonparametric regression model under monotonicity or convexity assumptions. To incorporate these constraints in the estimation procedure a modified version of a convolution type kernel estimator is considered. Some simulations are performed in order to evaluate empirically the amount by which this technique improves the estimation of the functional parameter in terms of quadratic error loss. The authors show that for a light computational price the observed improvements can be significant.