The mean-variance cardinality constrained portfolio optimization problem using a local search-based multi-objective evolutionary algorithm
The mean-variance cardinality constrained portfolio optimization problem using a local search-based multi-objective evolutionary algorithm
复制标题
DOI：
10.1007/s10489-017-0898-z
复制
发表时间：
2017-09
影响因子：
5.3
通讯作者：
Bili Chen;Yangbin Lin;Wenhua Zeng;Hang Xu;Defu Zhang
中科院分区：
文献类型：
--
作者：
Bili Chen;Yangbin Lin;Wenhua Zeng;Hang Xu;Defu Zhang
投资组合优化问题是金融学中的一个重要研究课题。这个问题的标准模型被称为马科维茨均值-方差模型，它有两个相互冲突的标准：预期收益和风险。本文考虑了一个更现实的同时包含基数约束和数量约束的投资组合优化问题，称为马科维茨均值-方差基数约束投资组合优化问题(MVCCPO问题)。我们扩展了一种基于多目标进化框架的算法，该框架结合了局部搜索模式和非支配排序。为了定量分析提出的算法的有效性，我们在涉及225个资产的公共可用数据集上将我们的算法与其他五种算法进行了比较。根据算法的基本操作和步骤，分别提出了边界约束处理策略、局部搜索策略、替换策略和最远候选法。通过仿真结果显示了该练习的成功。不同基数约束下的实验结果表明，该算法在邻近度和多样性方面均优于其他算法。此外，还研究了算法中使用的多样性维护策略，并利用扩散度量来评估所获得的非支配解的分布情况。文中还对算法的灵敏度进行了实验研究。
Portfolio optimization problem is an important research topic in finance. The standard model of this problem, called Markowitz mean-variance model, has two conflicting criteria: expected returns and risks. In this paper, we consider a more realistic portfolio optimization problem, including both cardinality and quantity constraints, which is called Markowitz mean-variance cardinality constrained portfolio optimization problem (MVCCPO problem). We extend an algorithm which is based on a multi-objective evolutionary framework incorporating a local search schema and non-dominated sorting. To quantitatively analyze the effectiveness of the proposed algorithm, we compared our algorithm with the other five algorithms on public available data sets involving up to 225 assets. Several modifications based on the fundamental operators and procedures of the algorithm, namely, the boundary constraint handling strategy, the local search schema, the replacement strategy and the farthest-candidate approach, are proposed one-by-one. Success of this exercise is displayed via simulation results. The experimental results with different cardinality constraints illustrate that the proposed algorithm outperforms the other algorithms in terms of proximity and diversity. In addition, the diversity maintenance strategy used in the algorithm is also studied in terms of a spread metric to evaluate the distribution of the obtained non-dominated solutions. The sensitivity of our algorithm has also been experimentally investigated in this paper.