Improving bandwidth selection methods by adding quantitative constraints (Q1775974)

In the context of kernel density estimation the authors propose to combine automatic bandwidth selection rules aiming at the minimization of a loss criterion with constraints on the search space derived from qualitative considerations. They focus on the cross-validation method of bandwidth and propose to derive constraints on bandwidths from reasonable bounds for the number of modes of the underling density. Theoretical properties of the chosen bounds are given. Results of simulation studies, which demonstrate improvement of the mean integrated squared errors of the kernel density estimation, are discussed.