Method for constructing Stieltjes classes for \(M\)-indeterminate probability distributions (Q1780524): Difference between revisions

Given a distribution function \(F\) with density \(f\), concentrated on the positive real line and with finite moments of all orders, assume that the moments of \(F\) do not identify it (\(F\) is \(M\)-indeterminate). The Stieltjes class of \(f\), defined by \[ S(f,h):=\{f_{\varepsilon}:=(1+\varepsilon f), \varepsilon\in\left[-1,1\right]\}, \] was introduced by the first author [J. Appl. Probab. 41A, 281--294 (2004; Zbl 1070.60012)]. All the densities in the class have the same moments, as a consequence of the assunption \(\int_0^{\infty} t^kf(t)h(t)dt=0\) for every natural \(k\). The index of dissimilarity in the Stieltjes class \(S(f,h)\) is defined by \( D_S:=\int_0^{\infty} | f(t)| \,h(t)\,dt. \) In Theorems A and D, the authors give a general method for constructing Stieltjes classes; it is based on a property of invariance under dilations, shifts and linear combinations (Lemma 1). The method is then applied to describe the \(M\)-indeterminacy of the power transformations of the lognormal, the inverse Gaussian and logistic distributions.