Title: Bandwidth-free pilot pre-smoothing for circular data
Authors: Jose E Chacon - Universidad de Extremadura (Spain) [presenting]
Carlos Tenreiro - University of Coimbra (Portugal)
Abstract: Least-squares cross validation (LSCV) yields a simple automatic bandwidth selection technique for kernel density estimation with circular data. Likelihood cross validation (LCV) represents another viable alternative, which in addition has shown a more efficient behaviour in practice. A smoothed cross validation (SCV) procedure is investigated, in which pilot pre-smoothing is employed to improve the performance of LSCV. However, contrary to the usual scenario, no pilot bandwidth or reference distribution is needed, so that the new method is fully data-driven. The practical performance of this SCV approach is compared with the traditional selectors in an extensive simulation study.