ポートフォリオ最適化のための進化アルゴリズム感応度測定。

抄録

Mean-Variance Portfolio optimization is highly sensitive to errors in assets means, when solved by Quadratic Programming (QP). We propose a simulation-based evaluation method for the sensitivity, applied to QP and the Evolutionary Algorithms (EA): Genetic Algorithm, Evolution Strategy, Particle Swarm Optimization and Differential Evolution. Comparisons between variants of EAs and QP are made based on assessing the performance, under multiple perturbed runs, of several ‘optimal’ portfolios. Computational experiments show that many individuals of EAs population outperform QP optimal solution.