A new class of kernels for nonparametric curve estimation (Q2366733)

A new class of variable kernels is introduced which, in the context of nonparametric curve estimation, have good convergence properties w.r.t. integrated mean square error. These kernels deform automatically near the boundary thus eliminating boundary bias. The kernels are closely related to the equivalent kernels of smoothing spline estimators, but have the advantage of depending on the smoothing parameter through a simple scaling operation.