No label defined (Q4328633)

Let \(f:\mathbb{R}^n\to\mathbb{R}\) be an unknown function. The following regression is investigated: \[ x_t=f(x_{t-1},\dots,x_{t-n})+\sigma\xi_t,\quad t=1,2,..., \] where \(\xi_t\) are independent random variables with zero mean value and unit variance \(\sigma>0\). The problem is to estimate the function \(f\), given \(\{x_t\}\). Given \(H>0\), let \(\tau=\inf \{N: B(N,h)\geq H\}\), where \(B(N,h)=\sum_{t=1}^N\delta(h,z,v_t)\), \(v_t=(x_{t-1},\dots,x_{t-n})\), and \(\delta(h,z,v)\) is a kernel of radius \(h\). In this paper, the kernel estimator \[ f^*_\tau(z)=H^{-1}\Biggl[\sum_{t=1}^{\tau-1} \delta(h,z,v_t)x_t+\alpha\delta(h,z,v_\tau)x_\tau\Biggr] \] is considered, where \(\alpha\) is defined by the equality \[ \sum_{t=1}^{\tau-1}\delta(h,z,v_t)+\alpha\delta(h,z,v_\tau)=H. \] The author states that, under some conditions, \(\mathbf P\{\tau<\infty\}=1\) and \[ {\mathbf E} |f_\tau(z)-f(z)|\leq \max_{||v-z||\leq h}|f(v)-f(z)|+\sigma/\sqrt H. \] The paper continues the studies by \textit{V.Z. Borisov} and \textit{M.G. Chistyakov} [Mat. Zamet. YAGU 6, No. 2, 3-8 (1999; Zbl 0944.62039)]. For the entire collection see [Zbl 0959.00045].