Asymptotic Uniqueness of the Non-Linear Regression Model Parameter with Least Squares Estimator

Abstract

In the paper the nonlinear regression model with continuous time and random noise, which is a local functional of strongly dependent stationary Gaussian random process, is considered. Sufficient conditions of asymptotic uniqueness of the least squares estimator of regression function parameters are obtained. This result is applied to the least squares estimator of amplitude and angular frequencies of harmonic oscillations sum observed on the background of given random noise. To obtain the main result limit theorems of random processes, weak convergence of a family of measures to the spectral measure of a regression function, etc were used. The novelty, compared with the known results in the theory of periodogram estimator in observation models on weakly dependent noise, is assuming that the random noise is a local functional of Gaussian strongly dependent stationary process. The result can be used in the proof of the asymptotic normality of the least squares estimator of nonlinear regression model parameters with the help of Brower fixed point theorem.