A large deviation result for the least squares estimators in nonlinear regression (Q689473)

Let \(\hat\theta_ n\) be the least squares estimator of a parameter \(\theta\in R^ k\) in a nonlinear regression model with dependent errors. An exponential inequality for the probability of large deviations of \(\hat\theta_ n\) from the true \(\theta\) is proved. In Theorem 1 it is required that the errors form a locally generalized Gaussian sequence. In Theorem 2 it is assumed that the errors are a multiplicative and adapted martingale difference sequence. The proved inequalities generalize results of \textit{B. L. S. Prakasa Rao} [see. Stat. Probab. Lett. 2, 139-142 (1984; Zbl 0544.62060), and J. Multivariate Anal. 14, 315-322 (1984; Zbl 0529.62053)] and of \textit{A. Sieders} and \textit{K. Dzhaparidze} [Ann. Stat. 15, No. 3, 1031-1049 (1987; Zbl 0661.62021)].