Selecting software projects to build a portfolio is a risky task. The Modern Portfolio Theory (MPT) has been proposed by many authors to support selecting the best group of projects that maximize the expected return x risk ratio of a portfolio. However, one major issue has to be resolved to allow the application of the theory when migrating its concepts to a project selection context, i.e, the way correlations between projects are calculated. We present a new technique to calculate them objectively, based on a risk scenario approach.