Smoothing dependent observations

We consider the problem of nonparametric regression when the error process has a continuous covariance structure. We show that smoothers are generally not consistent. If measurements from independent trajectories are available, smoothing the averaged trajectories can be detrimental asymptotically.