Trajectories of exchangeable sequences: Large and moderate deviations results (Q1273011)

Let \(P\) be a probability measure defined on the measurable space \((\Omega, {\mathcal F})\), \(\{P(\theta, \cdot)\}_{\theta\in\Theta}\) a family of probability measures also on \((\Omega, {\mathcal F})\), and \((\Theta, {\mathcal M}, m)\) a probability space with \(\Theta\) a first countable topological space, and \(P(\cdot, \cdot)\) a transition probability function on \(\Theta\times{\mathcal M}\). Let further \(\{Y_j\}_{j=1}^\infty\) be a sequence of r.v.'s defined on the measurable space \((\Omega,{\mathcal F})\) and taking values in the space \(({\mathbb R}^d,{\mathcal B}_{{\mathbb R}^d})\), \(d\geq 1\). Assume \(P\) is a mixture of probability measures \(\{P(\theta, \cdot)\}_{\theta\in\Theta}\) with the mixing measure \(m\), and that for each \(\theta\in\Theta\), the sequence \(\{Y_j\}_{j=1}^\infty\) is i.i.d. with respect to \(P_\theta(\cdot):=P(\theta, \cdot)\), with common distribution \(\pi_\theta=P_\theta\circ Y_1^{-1}\). Define \(S_0=0\), \(S_j=\sum_{i=1}^j Y_i\), \(j=1,2,3,\ldots\), and let \(s_n(t)\), \(t\in [0,1]\), denote the polygonal line in \({\mathbb R}^d\) determined by the points \((j/n,S_j/b_n)\), \(j=0,1,2,\ldots,n\), and \(\{b_n\}_{n=1}^\infty\) a positive real sequence. A large deviation result is proved for the sequence of probability measures \(\mu_n={\mathcal L}_P(s_n(\cdot))\) when \(b_n=n\). This result is a generalization of a result by \textit{A. A. Mogul'skij} [Theory Probab. Appl. 21, 300-315 (1976); translation from Teor. Veroyatn. Primen. 21, 309-323 (1976; Zbl 0366.60031)]. A moderate deviations result is proved for a sequence of probability measures \(\mu_n\) when \(b_n\) is such that \(b_n/n^{1/2}\to\infty\) and \(b_n/n\to 0\) as \(n\to\infty\).