Large deviations probabilities for recursive m-estimators (Q4364124)

Suppose that \(X_1, \dots, X_n\) is a random sample from a population with parameter \(\theta\) which is unknown but is known to lie in an interval of the form \([-A,A]\). Let \(\psi: R^2 \to R\) be a measurable function. Given an initial condition \(z_0\), define the recursive minimum-contrast estimators \(\{z_n\}\) by \(z_n= z_{n-1} +\psi (X_n, z_{n-1})/nJ\) for a normalization factor \(J>0\) depending on \(\psi\). This paper presents sets of hypotheses under which \(\lim_{n\to \infty} n^{-1} \log P[|z_n- \theta |\geq c]\), the limit of the Bahadur risks, exists and equals an explicit function of \(\theta\) and \(c>0\). An asymptotically optimal two-stage recurrent procedure is also derived.