Concavity and estimation (Q913397): Difference between revisions

Given a ``sample space'' U (a separable complete metric space) and an auxiliary set \(V=R^ p\) for some finite positive integer p, an estimation function \(h: U\times V\to [-\infty,+\infty)\) is considered. The special requirement is that h(x,\(\cdot)\) is concave for each \(x\in U.\) Let \(d(w,h,F)=E h(X,w)\), where X is an U-valued random element with distribution F. Given \(W\subset V\), the M-parameter M(W,h,F) is defined, roughly speaking, as a w maximizing d(w,h,F) with respect to \(w\in W\). To estimate M(W,h,F) the estimates \(M(W,h,F_ n)\) are used, where \(F_ n\) is the empirical distribution. Strong consistency and asymptotic normality of \(M(W,h,F_ n)\) is discussed.