scientific article; zbMATH DE number 1498457

From MaRDI portal
Publication:4499529





Let \(f\:{\mathbb R}^m\to{\mathbb R}\) be an unknown function. The following regression is under study: \[ y_s=f(x_s)+\sigma_s\xi_s,\quad s=1,2,\dots, \] where \(\{x_s\}\) are independent \({\mathbb R}^m\)-valued random variables distributed with density \(g(x)\). It is assumed that \({\mathbb E}\{\xi |{\mathcal F}_{s-1}^\xi,{\mathcal F}_t^x\}=0\) and \({\mathbb E}\{\xi_s^2 |{\mathcal F}_{s-1}^\xi,{\mathcal F}_t^x\}=1\) for every \(t\geq 1\), where the \(\sigma\)-algebra \({\mathcal F}_t^x\) is generated by the family \(\{x_1,\dots,x_t\}\) and \({\mathcal F}_s^\xi\) is generated by \(\{\xi_1,\dots,\xi_s\}\). The values \(\sigma_s\) are assumed to be bounded by some constant \(M_\sigma\). The problem is to estimate the function \(f\), given \(\{x_s\}\) and \(\{y_s\}\). Given \(H>0\), put \(\tau=\inf\{n: B(n,h)\geq H\}\), where \(B(n,h)=\sum_{s=1}^n\delta(h,z,x_s)\) and \(\delta(h,z,x)\) is a kernel of radius \(h\). In this article, the kernel estimator \[ f^*_\tau(z)=H^{-1}\Biggl[\sum_{s=1}^{\tau-1} \delta(h,z,x_s)y_s+\alpha\delta(h,z,x_\tau)y_\tau\Biggr] \] is considered, where \(\alpha\) is defined by the equality \[ \sum_{s=1}^{\tau-1}\delta(h,z,x_s)+\alpha\delta(h,z,x_\tau)=H. \] The authors show that \(\tau\) is finite with probability 1 whenever \(g(z)>0\) and establish the following estimate: \[ {\mathbb E}|f^{*}_\tau(z)-f(z)|\leq \max_{|x-z|\leq h}|f(z)-f(x)|+\sqrt{M_\sigma/H}. \] .











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 Q4499529)