Integrated square error of nonparametric estimators of regression function: The fixed design case (Q1200739): Difference between revisions

Let \(g\) be a regression function defined on the integral [0,1], and at the fixed points \(x_{n1},\dots,x_{nn}\) consider the observations \(Y_{ni}=g(x_{ni})+\varepsilon_{ni}\), \(i=1,\dots,n\), where the \(\varepsilon_{ni}\) are the error variables. This paper investigates the general linear estimate of \(g\) defined by \(\widehat g_ n(x)=\sum^ n_{i=1} W_{ni}(x)Y_{ni}\), \(x\in[0,1]\), and studies the asymptotic properties of the integrated square error \(I_ n=\int^ 1_ 0 \bigl(\widehat g_ n(x)-g(x)\bigr)^ 2 w(x)dx\), where \(w(x)\) is a weight function. Under some assumptions and using the method of \textit{P. Hall} [J. Multivariate. Anal. 14, 1-16 (1984; Zbl 0528.62028)], for \(n\to\infty\) the following results are obtained: \[ (I_ n-EI_ n) u^{-1/2}_ n\overset {d}\longrightarrow N(0,4\sigma_ 1), \] \[ I_ n-EI_ n\overset {p}\longrightarrow 0,\qquad I_ n-EI_ n\longrightarrow 0\quad\text{a.s.} \] {}.