One important concept in financial risk management is the diversification process of capital allocation. This paper proposes an evolutionary approach for the optimal diversification when making asset allocation using variable-size genetic relation algorithm (vs-GRA), whose main role is to model and evolve structures toward effective and diversified portfolios through its graph structure. Simulations using heterogeneous and globally located asset classes in the United States, Europe, and Asia show that the proposed scheme offers competitive economic advantages. © 2012 Institute of Electrical Engineers of Japan.