No label defined (Q4538797)

The authors investigate the nonlinear model \(X_j=g(j,\theta)+\epsilon_j,\;j=1,\dots,n\), where \(g(j,\theta),\;j\geq 1\), are non-random functions defined on the closure \(\Theta^c\) in \(\mathbb{R}^q\) of a bounded set \(\Theta\), and \(\xi_j,\;j\geq 1,\) are i.i.d. random variables. The \(l_{\alpha}\)-estimator \(\hat{\theta}_n=\hat{\theta}_n(X_1,\dots,X_n)\in\Theta^c\) for the parameter \(\theta\in\Theta\) from observations \(X_1,\dots,X_n\) is determined by the property \[ S_{\alpha}(\hat{\theta}_n)=\inf_{\tau\in\Theta^c}S_{\alpha}(\tau),\quad S_{\alpha}(\tau)=\sum_{j=1}^n |X_j-g(j,\tau)|^{\alpha}. \] These estimators are from the class of \(M\)-estimators [see \textit{P.J. Huber}, Robust statistics. (1981; Zbl 0536.62025)]. The most investigated estimators of this type are the least squares estimator \(l_2\), and the least moduli estimator \(l_1\) [for the corresponding references see the books: \textit{A. Pázman}, Nonlinear statistical models. (1993; Zbl 0808.62058); \textit{A.V. Ivanov}, Asymptotic theory of nonlinear regression. (1997; Zbl 0874.62070)]. The authors prove a theorem on asymptotic normality of the \(l_{\alpha}\)-estimator \(\hat{\theta}_n\) of the parameter \(\theta\) in the model for an arbitrary \(\alpha\in(1,2)\). The idea of the proof goes back to Huber's results. For linear regression models such a result was proved by \textit{A.E. Ronner} [Z. Wahrscheinlichkeitstheor. Verw. Geb. 66, 613-620 (1984; Zbl 0543.62025)].