Bandwidth selection for the local polynomial estimator under dependence: a simulation study (Q2488423)

A univariate nonparametric fixed regression model is considered with errors which follow an ARMA model. Local polynomial estimates (LPR) and generalization (GLPR), in which a data transformation is used to derive uncorrelated errors, are described. Seven procedures of bandwidth selection for these estimates are considered: different versions of cross-validation (including time-series CV and CV with prediction), asymptotic and exact plug-in and bootstrap methods. These procedures are compared via simulations.