No label defined (Q4940257)

Let \(D\) be the space of all càdlàg functions \(x=(x_{t},\;t\geq 0)\) (the Skorokhod space). Let \(G_{t}\) be the \(\sigma\)-field generated by all maps \(x\to x_{s}\) for \(s\leq t\) and \(G=\bigcup_{t\geq 0}G_{t}.\) Let \(P_1\) and \(P_2\) be two probability measures on the measurable space \((D,G)\), \({\mathcal F}_{t}=\bigcap_{s>t}G_{s}.\) Let \(P_{1}^{t}\) and \(P_{2}^{t}\) be restrictions of the measures \(P_{1}\) and \(P_{2}\) to the \(\sigma\)-field \({\mathcal F}_{t}.\) Let \(z_{t}=dP_{1}^{t}/P_{2}^{t}\) be the Radon-Nikodym derivative of the measure \(P_{1}^{t}\) with respect to the measure \(P_{2}\) for all \(t\in R_{+}.\) The authors denote by \(\zeta_{1}=\{\zeta_{1t},\;t\geq 0\}\) and \(\zeta_{2}=\{\zeta_{2t},\;t\geq 0\}\) the local density processes of the measures \(P_{1}\) and \(P_{2}\), resp., with respect to the measure \(Q=2^{-1}(P_{1}+P_{2})\) and consider the case \(P_{1} \ll P_{2}\) when \(z=\zeta_{1}/\zeta_{2}.\) They introduce the Hellinger integral \(H_{t}(\varepsilon)\) of order \(\varepsilon\) for the measures \(P_{1}^{t}\) and \(P_{2}^{t}\) by the equality \(H_{t}(\varepsilon)=H(\varepsilon,P_{1},P_{2})=E_{Q}Y_{t}(\varepsilon)\), where \(Y_{t}(\varepsilon)= \zeta_{1}^{\varepsilon}(t)\cdot\zeta_{2}^{\varepsilon}(t).\) They introduce the Hellinger process, too. In the case of locally absolutely continuous measures the form of the Hellinger process of order \(\varepsilon\) is obtained. On the basis of these results the authors prove large deviation theorems for the logarithm of the likelihood ratio. The rate of decrease for 1 st and 2 nd type error probabilities of Neyman-Pearson tests is investigated, too. Some examples are considered. Note, that the paper contains some misprints. For example, the indexes 1 and 2 in the probability measures \(P_1\) and \(P_2\) are missed.