In many cases, collusive agreements are formed by asymmetric firms and include only a subset of the firms active in the cartelized industry. This paper endogenizes the process of cartel formation in a numeric simulation model where firms differ in marginal costs and production technologies. The paper models the incentive to collude in a differentiated products Bertrand-oligopoly. Cartels are the outcomes of a dynamic formation game in mixed strategies. I find that the Nash-equilibrium of this complex game can be obtained efficiently by a Differential Evolution stochastic optimization algorithm. It turns out that large firms have a higher probability to collude than small firms. Since firms' characteristics evolve over time, the simulation is used to generate data of costs, prices, output-quantities, and profits. This data forms the basis for an evaluation of empirical methods used in the detection of cartels.