Locally adaptive smoothing splines combine features of variable kernel estimators and smoothing splines allowing for local adaptive fitting of a nonparametric regression function with splines. Basically, one presmooths the raw data with a local bandwidth kernel estimator, and then computes a global fit to the presmoothed data using a penalized likelihood. The resulting estimator is a locally adaptive smoothing spline easily computed for any point in the domain of the regression function from the values of the locally adaptive kernel estimator on a grid. We present some asymptotic properties of this estimator and study its finite sample behavior through simulations.
展开▼
