No label defined (Q4248580)

Let \(\{X_i; i\geq 1\}\) be independent and identically distributed (i.i.d.) random variables with probability density function \(\sigma^{-1}((x-\mu) /\sigma)\) with respect to the Lebesgue measure, where \(f\) is known and \(\theta= (\mu,\sigma)\) \((\sigma>0)\) is unknown. We consider estimating \(\mu\) under the loss function \(L(\theta,d)= \rho(d-\mu)\), where \(\rho(u)\) is nonincreasing for \(u<0\), nondecreasing for \(u>0\) and \(\rho(0)=0\). Let \[ M_{\min} =\min(m_-, m_+), \quad M_{\max}= \max(m_-,m_+), \] where \(m_-= \lim_{u\to-\infty} \rho(u)\) and \(m_+= \lim_{u\to \infty} \rho(u)\), which may be infinity. Let \(\delta (\mathbf {X})\) be an estimator based on \(\mathbf{X}= (X_1,\dots,X_n)\). Then the purpose of this paper is to find if there exists an estimator \(\delta(\mathbf {X})\) such that \[ E_\theta\biggl\{ L\bigl(\theta, \delta(\mathbf{X}) \bigr) \biggr\}\leq W,\tag{1} \] for all \(\theta\), where \(W(>0)\) is a given constant. Under a symmetric loss function \textit{E. L. Lehmann} [Notes on the theory of estimation. Univ. California Press (1951)] showed that if the sample size \(n\) is predetermined, then no estimators meet (1) for \(W<M_{\min} (=M_{\max})\). In Section 2 we shall show that the nonexistence is also true for an asymmetric loss function if \(W< M_{\min}\), but there does exist such an estimator if \(M_{\min} <W<M_{\max} < \infty\). However, we can not show if the result also holds when \(M_{\max}= \infty\). In Section 3 employing two-stage procedures by \textit{C. Stein} [Ann. Math. Statist 16, 243-258 (1945; Zbl 0060.30403)] we shall show that it is possible to construct an estimator meeting the requirement (1) for some distributions. Section 4 is devoted to the normal distribution and discusses the property of the proposed two-stage procedure.