scientific article; zbMATH DE number 1286002

From MaRDI portal
Publication:4242553





Let \(f\) be a positive function. The following regression is studied: \[ y_s=f(x_s)+\sigma_s\xi_s, \quad s=1,\ldots,N, \] where \(\xi_s\) are noncorrelated random variables with \(\text{\textbf{E}}\xi_s=0\) and \(\text{\textbf{D}}\xi_s=1\). The problem is to estimate the derivative of the function \(f\), given \(\{x_s\}\) and \(\{y_s\}\). To construct an estimator for \(f'\), the authors use the following kernel estimator for \(f\) [see \textit{V.Z. Borisov} and \textit{M.G. Chistyakov}, Mat. Zamet. YAGU 2, No.2, 11-21 (1995; Zbl 0870.62030)]: \[ f_N(h,z,x)=\sum_{s=1}^N\frac{\delta(h,z,x_s)}{B(N,h)}y_s, \] where \(B(N,h)=\sum_{s=1}^N\delta(h,z,x_s)\) is a normalizing coefficient and \(\delta(h,z,x)\) is a kernel. Taking an interval \(\Delta_N\) such that \(\Delta_N\to 0\) and \(\Delta_NB(N,h)\to\infty\) as \(N\to\infty\), the authors consider the following estimator: \[ f'_N(z,\Delta)=\frac{f_N(h,z+\Delta_N,x)-f_N(h,z,x)} {\Delta_N}. \] Necessary and sufficient conditions are given under which \(f'_N(z,\Delta)\to f'(z)\) in \(L_2\) as \(N\to\infty\). The convergence with probability 1 is also established.











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 Q4242553)