Asymptotic expansion for the distribution of the dispersion of the observation error in a nonlinear regression model (Q5905407)

A nonlinear regression model \(X_ j=g(j,\theta)+\varepsilon_ j\), \(j=1,\ldots,n\), with independent identically distributed random observation errors \(e_ j\), \(E \varepsilon_ j=0\), \(E \varepsilon^ 2_ j=\sigma^ 2>0\), is considered. Let \(L_ n(\theta)=\sum^ n_{j=1}[X_ j-g(j,\theta)]^ 2\); \(\hat\theta_ n\) is the least squares estimator of the regression function \(g(j,\theta)\) of the parameter \(\theta\in\Theta\subseteq R^ q\). The asymptotic expansion (a.e.) of the \(\sigma^ 2\) estimator \(\hat\sigma^ 2_ n=n^{-1}L_ n(\hat\theta_ n)\) and its distribution function under some regularity conditions on \(g(j,\theta)\) and \(\varepsilon_ j\) are obtained. The first two terms of those a.e. are given.