摘要
A portfolio optimisation problem involves allocation of investment to a number of different assets to maximize yield and minimize risk in a given investment period. The selected assets in a portfolio not only collectively contribute to its yield but also interactively define its risk as usually measured by a portfolio variance. In this paper we apply various techniques of multiobjective genetic algorithms to solve portfolio optimization with some realistic constraints, namely cardinality constraints, floor constraints and round-lot constraints. The algorithms experimented in this paper are Vector Evaluated Genetic Algorithm (VEGA), Fuzzy VEGA, Multiobjective Optimization Genetic Algorithm (MOGA) , Strength Pareto Evolutionary Algorithm 2nd version (SPEA2) and Non-Dominated Sorting Genetic Algorithm 2nd version (NSGA2). The results show that using fuzzy logic to combine optimization objectives of VEGA (in VEGAFuzl) for this problem does improve performances measured by Generation Distance (GD) defined by average distances of the last generation of population to the nearest members of the true Pareto front but its solutions tend to cluster around a few points. MOGA and SPEA2 use some diversification algorithms and they perform better in terms of finding diverse solutions around Pareto front. SPEA2 performs the best even for comparatively small number of generations. NSGA2 performs closed to that of SPEA2 in GD but poor in distribution.
摘要译文
投资组合优化问题涉及的投资分配到多个不同的资产，以最大限度地提高产量和减少在给定的投资期限风险。在投资组合的资产选择，不仅共同推动其收益率也交互定义其风险通常是由投资组合的方差衡量。在本文中，我们运用了多目标遗传算法的各种技术来解决投资组合优化与一些现实的约束条件，即基数限制，地板约束和圆很大的制约。试验本文中的算法是向量评估遗传算法（VEGA），模糊VEGA，多目标优化遗传算法（MOGA）强度帕累托进化算法第二版（SPEA2）和非支配排序遗传算法第二版（NSGA2）。置的通过产生距离（GD）由上一代人口的至真帕累托前的最近的成员，但其解决方案的平均距离定义往往集中围绕几点。MOGA和SPEA2使用一些多元化的算法，他们在寻找周围的帕累托前不同的解决方案方面有更好的表现。SPEA2表现最好即使是比较小的数代人。 NSGA2执行封闭到SPEA2在GD，但贫困人口的分布。
Skolpadungket, P.; Dahal, K.; Harnpornchai, N.. Portfolio optimization using multi-obj ective genetic algorithms[C]//Evolutionary Computation, 2007. CEC 2007. IEEE Congress on, Singapore, 25-28 Sept. 2007, SG: IEEE, 2007: 516-523