On the asymptotic formula for the probability of a type I error of mixture type power one tests (Q1084817)

Let \(X,X_ 1,X_ 2,..\). be i.i.d. real valued with distribution \(f_{\theta}d\mu\), in which \(f_{\theta}(x)=\exp (\theta x-\psi (\theta))\), \(\theta \in \Omega =open\) interval. Wlog assume \(0=\psi (0)=\psi '(0)\). Let F be the distribution function of a prior on \(\Omega\) and define \[ L_ n(F)=\int \prod^{n}_{i=1}f_{\theta}(X_ i)/f_ 0(X_ i)dF(\theta). \] For testing (with an open-ended test) \(H_ 0: \theta =0\) versus \(H_ 1: \theta \neq 0\) define the stopping time \(T=T(B,F)\) to be the least \(n\geq 1\) such that \(L_ n(F)\geq B\), with the acceptance of \(H_ 0\) iff \(T=\infty\). About the rate at which the significance level goes to 0 as \(B\to \infty\) it is known that \[ BP_ 0\{T(B,F)<\infty \}\to \int_{\Omega}\int^{\infty}_{0}e^{-x} dH_{\theta}(x)dF(\theta), \] in which \(dH_{\theta}\) is the asymptotic distribution of the excess over a large stopping bound of a random walk with steps \(\theta X_ i-\psi (\theta)\). For this result it is assumed that F has a positive continuous density with respect to Lebesgue measure on \(\Omega\) and that \(\theta\) X-\(\psi\) (\(\theta)\) is nonlattice for every \(\theta\in \Omega\) if X has distribution \(f_ 0d\mu.\) The present paper extends this result in two ways. First, the same result is proved assuming on F only that \(F(\{0\})=0\) and that the set of \(\theta\in \Omega\) for which \(\theta\) X-\(\psi\) (\(\theta)\) has lattice distribution (when X has distribution \(f_ 0d\mu)\) has F-probability 0. Second, under the condition that X is strongly nonlattice under every distribution \(f_{\theta}d\mu\), \(\theta\in \Omega\), it is shown that for every [a,b]\(\subset \Omega\) the convergence as \(B\to \infty\) is uniform in \(F\in S[a,b]=the\) family of distributions F whose support is contained in [a,b].