LAN in Problems of Nonparametric Estimation of Functions and Lower Bounds for Quadratic Risks (Q4713745)

Let \(\{X^ \epsilon,B^ \epsilon,P_ \theta^ \epsilon,\theta\in\Theta\}\) be a family of statistical experiments, where \(\epsilon\to 0\), \(\theta\in L_ 2(T)\), \(T\) a measurable set in \(R\), and \(B^ \epsilon\) a measurable map \(X^ \epsilon\to L_ 2(T)\). Let \(\theta^*(t,x^ \epsilon)\), \(t\in T\), be some estimator of the parameter \(\theta\) based on \(x^ \epsilon\in X^ \epsilon\). The risk of the estimator is given by the following expression: \[ R_ \epsilon[\theta^*,B_ r(\theta)]=\sup_{\theta\in B_ r(\theta_ 0)} E_ \theta^ \epsilon \int_ T[\theta(t)-\theta^*(t,x^ \epsilon)]^ 2 dt, \] where \(E_ \theta^ \epsilon\) is the expectation with respect to the measure \(P_ \theta^ \epsilon\), \(B_ r(\theta_ 0)\in\Theta\), and \(B_ r(\theta_ 0)\to\theta_ 0\in\Theta\) for \(r\to 0\). A lower bound for \(R_ \epsilon\) is given.