Catastrophes produce rare and highly correlated insurance claims, which depend on the amount of coverage at different locations. A joint probability distribution of these claims is analytically intractable. The most promising approach for estimating total claims for a particular combination of decision variables involves geographically explicit simulations of catastrophes. The straightforward use of catastrophe models runs quickly into infinite "if--then" evaluations. The aim of this paper is to develop a framework allowing for the use of Monte Carlo simulation of catastrophes to aid decision making on designing optimal catastrophic risk portfolios. A dynamic optimization model is discussed. Connections between ruin probability and nonsmooth, in particular concave, risk functions are established. Nonsmooth adaptive Monte Carlo optimization is proposed.