This paper deals with the problem of attaining a set of targets (goals) by means of a set of instruments (subgoals) when the relation between the two groups of variables can be expressed with a linear system of stochastic equations. The objective function consists of the maximization of the probability that a realization (in terms of target variables) will lie in a confidence region of predetermined size. Under suitable normality assumptions this problem is amenable to a quadratic programming formulation.