基于链路预测对A股市场潜在影响力股票的分析

发布时间：2018-05-10 17:20

本文选题：股票网络市场 + 阈值法　；参考：《江苏大学》2017年硕士论文

【摘要】：近年,大量的研究人员主要是通过复杂网络的拓扑结构以及中心性度量的方法来描绘复杂网络中的基本特性。而在网络信息化社会中,复杂网络早已被广泛的利用到各行各业中,其中金融证券市场便是复杂网络的一个重要的讨论方向,使得复杂网络中的链路预测被提升为研究主体。在金融证券市场里股票价格的大幅波动会产生较大的风险。这不仅打破市场带来的均衡价值,甚至可能促使市场机能的彻底丧失。因此本论文将利用相关系数法与阈值法来构建股票网络市场,运用局部结构中心性来分析A股市中潜在影响力的股票,随后结合链路预测中相似性指标来衡量潜在影响力股票在未来相互连接的可能性程度。本论文主要工作是:首先主要是从拓扑结构中心性来衡量节点重要性的影响程度。其中通过比较了局部度量和全局度量的中心值,阐述其优缺点。即本论文将运用局部结构中心法来分析A股市场中股票的影响力,其中它的精确性远远高于局部度量,而计算复杂性远远低于全局度量。其次,收集每日A股股票的收盘价格数据,来做股票价格的对数相关系数计算,从而简单粗略地反映股票之间相关性的强弱。随之引入度量矩阵和阈值法来分析股票网络市场模型,并得出本论文研究股票网络市场模型的阈值为0.89较为合适。通过对阈值取值不同时,分别运用局部结构中心法来分析股票网络市场中潜在影响力的股票。最后,论文主要以股票节点之间的相似性越大连接可能性就越大为研究前提条件,以AUC为评价标准并利用控制变量法来比较局部信息相似性,分析出资源配置(RA)相似性指标为最佳理想的相似性指标。基于RA指标与局部路径相似性指标中网络的动态演化,可知最优相似性指标是在不断的发生变化。研究表明,通过对具体的网络形态利用RA指标与Katz指标来对A股中股票进行数值化分析,A股股票与有色金属类潜在影响力的股票之间可能发生连接性,以此使得投资者达到规避风险效果。
[Abstract]:In recent years, a large number of researchers describe the basic characteristics of complex networks mainly through the topological structure of complex networks and the method of central measurement. In the network information society, the complex network has been widely used in various industries, in which the financial security market is an important discussion direction of the complex network, which makes the link prediction in the complex network to be promoted to the research subject. In the financial securities market, large fluctuations in stock prices will have greater risks. This not only breaks the equilibrium value of the market, it may even cause the complete loss of market function. Therefore, this paper will use the correlation coefficient method and threshold method to construct the stock network market, and use the local structure centrality to analyze the potentially influential stocks in the A-share market. Then the similarity index in link prediction is used to measure the possibility of potential impact stocks to be connected in the future. The main work of this thesis is as follows: firstly, the influence degree of node importance is measured from topology centrality. The central values of local and global metrics are compared, and their advantages and disadvantages are expounded. That is to say, this paper will use the method of local structure center to analyze the influence of stock in A-share market, in which its accuracy is far higher than that of local measurement, and the computational complexity is far lower than the global metric. Secondly, the daily closing price data of A shares are collected to calculate the logarithmic correlation coefficient of stock prices, so as to reflect the correlation between stocks roughly. Then the measurement matrix and threshold method are introduced to analyze the stock network market model, and it is concluded that the threshold value of this paper is 0.89. The local structure center method is used to analyze the potentially influential stocks in the stock network market by taking different threshold values. Finally, the paper mainly takes the similarity between stock nodes as the premise of the study, taking AUC as the evaluation standard and using the control variable method to compare the similarity of local information. The similarity index of resource allocation is the best ideal similarity index. Based on the dynamic evolution of the network between the RA index and the local path similarity index, it can be seen that the optimal similarity index is constantly changing. The results show that by using RA index and Katz index to numerically analyze the A-share stock, the connection between A-share stock and non-ferrous metal potentially influential stock may occur. In this way, investors to achieve risk-aversion effect.
【学位授予单位】：江苏大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F832.51

【参考文献】