scientific article; zbMATH DE number 1337346

Publication date 15 September 1999

projection estimators random fields statistics rate of convergence

Nonparametric regression and quantile regression (62G08) Asymptotic properties of nonparametric inference (62G20) General nonlinear regression (62J02)

Review text

For regression models of the form \[ (i)\;x_j^n=a(j/n)+\xi_j^n,\;\;(ii)\;x_{ij}^{nm}=a(i/n,j/n)+\xi_{ij}^{nm},\;\;(iii)\;x_j^n(t)=a(j/n,t)+\xi_j^n(t), \] where \(x\) is the result of observations, \(a\) is an unknown function, \(\xi_{i,j}\) are independent errors of observations with \(E\xi_{ij}=0\), \(E(\xi_{ij})^2=\sigma^2\) (\(\xi_i(t)\) are independent random processes), projection estimators for \(a\) are considered. It is assumed that \( a(\cdot)=\sum_{i=1}^\infty a_j\phi_j(\cdot), \) where \(\{\phi_j(\cdot)\}_{j=1}^\infty\) is a fixed basis in \(L_2\) and \[ (1)\qquad \sum_j a_j^2(1+j^{2\alpha})<L<\infty. \] Estimators in (ii) are \(a_N^{nm}(s,t)=\sum_{l=1}^N a_l^{nm}\phi_l(s,t)\), where \[ a_l^{nm} = (nm)^{-1}\sum_{i=0}^n\sum_{j=0}^m x_{ij}^{nm} \phi_l(i/n,j/m). \] It is shown that for a class \(K\) of functions \(a\) such that (1) holds, \(\sup{}a(s,t){}<C\) for all \[ \phi_j(t,s)-\phi_j(t',s'){}\leq C_{j}\sqrt{(t-t')^2+(s-s')^2}, \] the relation \[ \lim_{n,m\to\infty}\sup_{a\in K} E\left( n^2m^2/\sqrt{n^2+m^2} \right)^{2\alpha/(2\alpha+1)} \|a_N^{nm}-a\|_{L_2}<\infty\;\text{as} N\sim \left( nm/\sqrt{n^2+m^2} \right)^{1/(2\alpha+3)} \] holds true. Analogous results are obtained for the models (i) and (iii).

Recommendations

Cited in

(15)

This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4264017)