Large deviation estimates for exceedance times of perpetuity sequences and their dual processes (Q504245)

The authors are concerned with path properties of perpetuity sequences \((Y_n)_{n \in \mathbb{N}}\) of the form \(Y_n = B_1 + A_1 B_2 + \dots + (A_1 \times \dots \times A_{n-1}) B_n\), where \((A_i, B_i)_{i \in \mathbb{N}}\) constitutes a sequence of stochastically independent and identically distributed random variables taking values in \((0, \infty) \times \mathbb{R}\). The authors' primary objective is to study the asymptotic distribution of \[ T_u = \frac{1}{\log(u)} \inf\{n: Y_n > u\} \] as \(u \to \infty\). They first prove that, conditionally to \(T_u < \infty\), \(T_u\) converges in probability to some positive constant \(\rho\) as \(u \to \infty\). Then, they derive large deviation results for \(\mathbb{P}(T_u \leq \tau)\) as \(u \to \infty\). It turns out that there exist three different asymptotic regimes, namely (i) \(\tau < \rho\), (ii) \(\tau = \rho\), and (iii) \(\tau > \rho\). While the original process \((Y_n)_{n \in \mathbb{N}}\) can be regarded as a backward stochastic process, also exceedance probabilities for forward recursive sequences are analyzed, by making use of duality arguments.