摘要
以复杂网络视角,采用自适应仿射传播聚类算法,从沪深300指数中选择分散化程度高的聚类中心作为标的,同时对比市值排序选股和权重排序选股,构建二次指数跟踪最优化模型,并进行实证分析与稳健性检验。研究发现:复杂网络聚类选股方法只需50只以内股票,即可实现较高精度的沪深300指数跟踪;当股票数量选择31只时,AAP聚类的误差相对较小且股票数目适中,样本内31只股票的日均跟踪误差为6.35×10^(-5),分别比市值排序和权重排序降低了36.50%和24.85%,样本外31只股票的日均跟踪误差为3.10×10^(-4),比其余两种方法分别降低了45.33%和38.74%;在稳定性检验中,当股票数介于5~35只之间时,AAP聚类选股进行指数跟踪优于其他两种选股的可能性超过80%,表明跟踪结果具有良好的稳健性。
From the perspective of complex networks,we use the adaptive affine propagation clustering algorithm to select the cluster center stock with high degree of decentralization from the CSI 300 Index as underlying assets and contrast with the methods of market capitalization sorters and weighted stock selection. Then we propose the construction of quadratic exponential tracking optimization model Index tracking model and make empirical analysis and robustness test. The empirical results show that,within 50 stocks of AAP cluster centers could achieve high precision replication on CSI 300 index. When 31 stocks are selected,the error of AAP clustering is relatively small and the number of shares is moderate. The daily average tracking error of 31 stocks inside the sample is 6. 35× 10^（-5）,36. 50% and 24. 85% respectively lower than market value order and weight sorting. The daily average tracking error of 31 stocks outside the sample is 3. 10 × 10^（-4）,45. 33% and 38. 74% respectively lower than the other two. Besides,in the stability test,AAP clustering is used to selected stocks to track market index,when the stock number is between 5 and 35,better than the other two kinds of stock selection possibility of more than 80%,indicating the results are equipped with good robustness.
出处
《经济问题》
CSSCI
北大核心
2018年第2期35-42,56,共9页
On Economic Problems
基金
国家自然科学基金项目(71101068
71171109)
中央高校基本科研业务费专项资金资助(011814380027)
江苏省自然基金面上项目(BK20161398)
江苏省金融工程重点实验室课题(NSK2015-09)
关键词
复杂网络
聚类选股
指数跟踪
自适应仿射传播聚类
日均跟踪误差
social network
stock clustering selection
index tracking
adaptive affinity propagation clustering
average tracking error
