New bounds for the probability that at least \(k\)-out-of-\(n\) events occur with unimodal distributions (Q2357784)

The ``\(k\)-out-of-\(n\) probability'' that out of \(n\) random events exactly \(k\) occur can be given by elementary closed formulae which are computationally expensive for large parameters \(n, k\). Approximations and bounds are well known as Bonferroni inequalities. The present paper approaches the problem via a transformation into a linear program. If \(\xi\) denotes the number of events occurring and \(p_i := \mathrm{Prob}(\xi=i)\), it is assumed that the sequence \(p_i,i=1\dots n,\) is unimodal and that for some \(m < n\) the binomial moments of the sequence are known. Given this, a linear program is developed to determine bounds for the probability that ``at least \(k\)-out-of-\(n\)'' events occur. The dual program formulation produces algorithmic solutions of this bounding procedure. The authors provide an application of the method to finance and compare their approach which one that utilizes shape information, i.e., it works with the help of unimodality and other bounds that do not incorporate such knowledge into the computations. Some special cases are investigated in detail.