No label defined (Q4677124)

The authors deal with a nonlinear regression model with continuous time and long-range dependent noise \[ X(t)=g(t,\theta)+\varepsilon(t),\quad t\in[0,T], \] where \(g(t,\theta)\colon[0,\infty)\times\Theta^c\to\mathbb R^1,\Theta^c\) is the closure in \(\mathbb R^q\) of an open set \(\Theta\subset\mathbb R^q\), and \(\varepsilon(t),t\in\mathbb R^1,\) is a real measurable mean-square continuous stationary Gaussian stochastic process with \(E\varepsilon(t)=0\) and covariance function \(E\varepsilon(0)\varepsilon(t)=B(t)=L(t)/| f|^{\alpha}\), \(0<\alpha<1,\) \(L(t)\) is a slowly varying at infinity function. The \(L_p\)-estimate of the unknown parameter \(\theta\in\Theta\) based on observations \(X(t),t\in [0,T],\) is a random vector \(\widehat\theta_T=\widehat\theta_T(X(t)\), \(t\in [0,T])\in\Theta^c\), that gives a solution to the problem \[ Q_{p,T}(\widehat\theta_T)=\inf_{\tau\in\theta_T}Q_{p,T}(\tau)=\int_0^T | X(t)-g(t,\tau)|^p\,dt. \] The least squares estimates and least module estimates for the proposed model were investigated by \textit{A. V. Ivanov} [Asymptotic theory of nonlinear regression. (1997; Zbl 0874.62070)]. In this paper, \(L_p\)-estimates are investigated for \(1<p<2\). The authors prove that under some conditions the normalized \(L_p\)-estimate \(\widehat{u}_T\) has an symptotically normal distribution: \[ \widehat{u}_T{\underset{T\to\infty}\Longrightarrow} N(0,\Lambda(\theta)\sigma(\theta)\Lambda(\theta)), \] where \(\Lambda(\theta)\) and \(\sigma(\theta)\) are matrices determined by derivatives of the function \(g(t,\tau)\).