Conditions of asymptotic efficiency of recursion estimates of the shift parameter (Q1103987)

Let \(Y_ 1,Y_ 2,...,Y_ n\) be i.i.d. random variables with a density \(p(y-\theta)\), where \(\theta \in R\) is an unknown parameter and P has a finite Fisher information \(I(p).\) The authors study recursive estimates \(\theta^*_ n\), which are asymptotically efficient. The sequence \(\theta^*_{n+1}=\phi_{n+1}(\theta^*_ n,Y_{n+1})\) of estimators suggested by the authors satisfies the following relations \(\forall y:\) \[ P_{\theta}\{(nI(p))^{1/2}(\theta^*_ n- \theta)<y\}\to \Phi (y), \] \[ M_{\theta}[f((nI(p))^{1/2}(\theta^*_ n-\theta))]\to \int_{R^ 1}f(y)d\Phi(y), \] where f is a continuous function and for some \(K>0\): \(\sup_{y}| f(y)| /(1+| y|^ K)<\infty\). This theorem generalizes a result of \textit{M. B. Nevel'son} [Theory Probab. Appl. 25, 569-579 (1981); translation from Teor. Veroyatn. Primen. 25, 577-587 (1980; Zbl 0444.62041)].