The finite-time ruin probability for an inhomogeneous renewal risk model (Q2358483): Difference between revisions

Consider a renewal type risk process \[ R(t) = x + c t - \sum_{i=1}^{N_t} Z_i\;, \] where \(N_t = \sup\{n \geq 0: \theta_1 + \cdots + \theta_n \leq t\}\) for some independent variables \(\{\theta_k\}\). The variables \(\{Z_i\}\) are iid, and \(\{Z_i\}\) and \(\{\theta_k\}\) are independent. It is assumed that \(\{\theta_k\}\) are uniform integrable, \(\sum_{i=1}^\infty \mathrm{Var}[\theta_i]/i^2<\infty\), and \(\lim_{n \to \infty} n^{-1} \sum_{i=1}^n E[\theta_i] = \lambda^{-1} \in (0,\infty)\). The quantity of interest is the finite-time ruin probability \(\psi(x;t) = P[\inf\{ R(s): 0 \leq s \leq t\} < 0]\). We let \(\mu = \lim_{t \to \infty} E[R(t)]/t\) and assume that \(\mu > 0\). The main result is the asymptotic formula for some subclass of the subexponential distributions for \(Z\) \[ {\psi(x;t) \over {1 \over \mu} \int_x^{x + \mu \lambda(t)} (1-F_Z(v))\;d v} \to 1 \] uniformly in \([T,\infty)\), where \(\lambda(t) = E[N_t]\), and \(T\) is such that \(\lambda(T) > 0\). In order to prove the result, the two inequalities \[ \liminf_{x \to \infty} \inf_{t \geq T} {\psi(x;t) \over {1 \over \mu} \int_x^{x + \mu \lambda(t)} (1-F_Z(v))\;d v} \geq 1-\epsilon\;, \] and \[ \limsup_{x \to \infty} \sup_{t \geq T} {\psi(x;t) \over {1 \over \mu} \int_x^{x + \mu \lambda(t)} (1-F_Z(v))\;d v} \leq 1+\epsilon \] are verified. The first of these two inequalities holds for the larger class of long-tailed distributions. The main idea of the proof is to condition on the number of claims until time \(t\) and the probability that the sum of claims exceeds a fixed level. In this way, the claim sizes and the claim arrivals can be split because they are independent.