Measuring operational risk using a mean scaled individual risk model (Q1826792)

The paper focuses on sums of \(n\) independent random variables \[ X=\sum_{i=1}^nX_i \] where \(X_i=\sum_{j=1}^{N_i}Z_{i,j}\) (\(i=1,\ldots, n\)), and the \(Z_{i,j}\)'s are \(iid\) as \(Z_i\), \(N_i\) is a random variable assuming nonnegative integer values, with \(N_i\) independent of the \(Z_{i,j}\)'s. The above finite sums \(X\) play a primary role in actuarial field as well as in risk management. In fact, in the first case, according to different asumptions about \(N_i\) (either Bernouili or Poisson distributions), \(X\) is linked to either individual or collective models of risk theory. Moreover in the managerial framework \(X\) is linked to operational risk models. On the basis of some results by the author himself already known in literature, sufficient conditions are provided, in order to guarantee that a finite mean scaled sum of compound random sums is a mean scaled sum of Gamma distributed components. Relating to the numerical evaluation of the distribution of \(X\), the author presents a combined approach, aimed to a unified discrete-continuous approximation method, based on the use of the cumulant generation functions, characterised by integral equations. Moreover an approximation method for generalized pseudo compound Poisson distributions is presented. Finally the results are applied to the construction of a mean scaled operational risk model.