Simple approximations for the ruin probability in the risk model with stochastic premiums and a constant dividend strategy (Q2218140)

A risk model \[ X_t = x + \sum_{i=1}^{\bar N_t} \bar Y_i - \sum_{j=1}^{N_t} Y_j - d t \] with two-sided jumps and dividend payments at constant rate \(d > 0\) is considered. The jump sizes are iid positive variables and the number of jumps are Poisson processes. All the variables are independent and a net profit condition \(\mathbb{E}[X_t - x] > 0\) is assumed. The ruin probability \(\psi(x) = \mathbb{P}[\inf_{t > 0} X_t < 0]\) is approximated by a de Vylder type approximation. That is, the ruin probability is approximated by the ruin probability of a model with exponentially distributed jump sizes \(Y_i\) and \(\bar Y_i\). For a first approximation the five possible parameters are chosen such that the first five moments of \((X_t - x)\) coincide. A problem may arise that some of the Parameters may become negative. For a second approximation, three moments are matched under the constraint that \(\mathbb{E}[\bar N_t]/\mathbb{E}[N_t]\) and \(\mathbb{E}[\bar Y_j]/\mathbb{E}[Y_j]\) are the same for the original and the approximative model. For exponentially distributed jump sizes, an explicit formula for the ruin probability can be obtained. Numerical examples for Gamma, hyperexponentially and Pareto distributed claim sizes show that the approximations work well for light-tailed jumps and less well in the Pareto case. A problem of the model is that the insurer will pay dividends even if it causes ruin. Thus, it would be better to interpret \(d\) as the rate of the administration costs. A variant of the model could be to allow \(d \le 0\). That is, in addition to the lump sum premia one adds a premium rate. The main difference would then be that the boundary condition \(\psi(0) = 1\) does not hold any more.