scientific article; zbMATH DE number 4028524

From MaRDI portal
Publication:3769698





[For the entire collection see Zbl 0619.00019.] Let \(P_ n\) and \(Q_ n\) be probability measures on measurable spaces \((\Omega_ n,{\mathcal F}_ n)\), \(n\geq 1\). The sequence \(\{P_ n\}\) is said to be contiguous with respect to \(\{Q_ n\}\) \((P_ n\Delta Q_ n)\), if \(\lim_{n}Q_ n(A_ n)=0\) implies \(\lim_{n}P_ n(A_ n)=0\) for any sets \(A_ n\in {\mathcal F}_ n\). It is known that \[ P_ n\Delta Q_ n\text{ iff } \lim_{n}P_ n'(\Omega_ n)=1\text{ and } \overline{\lim}_{c\to \infty}\overline{\lim}_{n}P_ n(X_ n>c)=0, \] where \(P_ n=P_ n'+P_ n\), \(P_ n'\ll Q_ n\), \(P_ n\perp Q_ n\) and \(X_ n=dP_ n'/dQ_ n\) [cf. \textit{G. G. Roussas}, Contiguity of probability measures. Some applications in statistics. (1972; Zbl 0265.60003)]. The main theorem of the paper states that if \(\{\phi_{\alpha}: R_+\to R_+\); \(0<\alpha <1\}\) is a family of measurable functions satisfying the conditions: \(\lim_{\alpha \uparrow 1}\phi_{\alpha}(x)=x\) uniformly on every finite interval, \(\lim_{x\to \infty}x^{- 1}\phi_{\alpha}(x)=0\) and \(\phi_{\alpha}(x)\) is a concave function of x for every fixed \(\alpha\in (0,1)\), then \[ P_ n\Delta Q_ n\quad iff\quad \underline{\lim}_{\alpha \uparrow 1}\underline{\lim}_{n}E_{Q_ n}\phi_{\alpha}(X_ n)=1. \] Basing on this result, some other criteria for contiguity of probability measures on measurable spaces with filtrations are established. Moreover, a necessary and sufficient condition for the uniform integrability of the Doleans-Dade exponential \({\mathcal E}(M)=\exp \{M-<M>\}\) is given where M is a local martingale with \(M_ 0=0\), namely: \[ \lim_{\alpha \uparrow 1}(1-\alpha)E[({\mathcal E}(M))^{\alpha}\circ <M>]_{\infty}=0. \]











This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3769698)