Asymptotic properties of nonparametric curve estimates (Q1066585): Difference between revisions

We consider the regression model \(Y_ i=g(t_ i)+\epsilon_ i\), \(1\leq i\leq n\), with nonrandom design points \(0\leq t_ 1\leq...\leq t_ n\leq 1\). In case no parametric model for the regression function g(.) is available and that only the smoothness of \(g(.)\) is known one can find several special estimators for g in the literature. Most of them are linear in the data \((Y_ j).\) In this paper we study a broad class of linear estimators \(\hat g \)(including kernel estimators, smoothed regressograms etc.) and evaluate their asymptotic behaviour. Under appropriate conditions on the r.v.'s \((\epsilon_ j)\), the design points \((t_ j)\) and the underlying approximation operator we prove local and global a.s. limit theorems, giving the exact speed of convergence of \(\hat g\) to g. For example a LIL for \((\hat g(t_ 0)-E(\hat g(t_ 0)))\) is given. Furthermore we derive the limit distribution for \(\| \hat g(.)- g(.)\|_{\infty}\) for estimators which behave asymptotically like kernel estimators. Most of the particular estimators of interest are of this type.