【作者】 马丽；

【导师】 扈文秀；

【作者基本信息】 西安理工大学 ， 金融学， 2019， 硕士

【摘要】 股票价格操纵行为给股票市场带来了巨大的危害,一方面会使股票价格偏离股票真实价值,严重损害普通投资者利益,削弱中小投资者的参与热情和投资信心;另一方面,严重制约着市场资源的有效配置,使股票市场逐渐萎缩,最终必然会给一国资本市场造成巨大损失。此外,近年来,股价操纵行为案件不断频发,且多数属于恶性操纵行为事件。虽然相关监管机构也在不断努力完善监管制度,但是随着监管力度的不断加强,传统的长线操纵模式逐渐消失,短线操纵模式不断成为主流。操纵时长由以往的1年或者更长时间变成了现在的1-2天,整个操纵过程极其短暂,基本在1-2天内就全部结束,这样就对监管机构的迅速反应能力构成了挑战和威胁。现有的股价操纵行为判别及预警指标基本均是金融交易指标(如:日收益率、波动率、换手率等)和财务指标。然而,这些指标只能在一个交易日结束之后,才可以计算得到,所以已有的判别方法存在严重的滞后性,根本不能即时甄别股市中的操纵行为,只适用于传统的长线操纵行为,并不适用于目前主流的短线操纵行为。鉴于此,本文在现有研究成果的基础上,从复杂网络的视角入手,主要从以下四个方面进行了研究:(1)基于被中国证券监督管理委员会认定在2015年发生的沪深A股市场交易型股票操纵行为案件为样本,运用日内分笔tick交易数据,以股票买卖双方的委托报单ID为节点,以买卖双方委托报单是否成交为连线构建股票交易网络。(2)构建股票交易型操纵行为判别模型。首先,通过逐步回归法筛选获得网络密度、节点数量、连线数量等13项能够刻画被实施操纵的股票在操纵期和非操纵期具有显著差异的网络参数;然后,通过因子分析法提取出5个主要的解释因子;最后,以所提取的主要因子为基础构建了股票交易型操纵行为判别模型。(3)对前文所构建的股票交易网络进行社团结构划分。为了找到隐藏于网络拓扑中的社团结构,提高股价操纵行为预警的精度。本文进一步从社团结构的角度,采用Louvain算法、VOS Clustering算法和GN算法对所构建的股票交易网络进行划分,然后对比哪一种社团划分方法效果更好。最终选用了 5只不同行业的股票进行了验证。(4)对股票交易型操纵行为预警系统设计进行了构想。针对当下操纵周期越来越短,操纵手法越来越隐蔽的现状,本文选用灵敏性较高的复杂网络参数和社团结构参数作为预警指标,利用所选参数构建了股票交易型操纵行为预警指数,并根据综合指标得分对股价操纵行为预警划分了等级。最终用一个案例进行了验证。通过本文一步步地研究与分析,最终,主要得到了以下三个结论:(1)利用复杂网络参数因子构建的股票交易型操纵行为判别模型具有较高的准确率。实证研究结果表明,本文构建的判别模型样本内检验总体准确率为86.%%,样本外检验总体准确率为86.5%。本文研究结果可为证券监管部门高效识别股票交易型操纵行为,合理分配监管资源,有效打击市场操纵行为提供技术支持。(2)在对股票交易网络进行社团划分时,选用了适合划分本文所构建网络的算法。VOS Clustering算法、Louvain算法和GN算法这几种划分效果相比,首先,VOS Clustering是最优的;其次,多数情况下Louvain算法的划分结果是优于GN算法的划分结果。(3)节点数的环比增长率、组元数的环比增长率、最大组元数环比增长率、社团数环比增长率、网络密度和最大组元规模比例可以作为股票交易型操纵行为预警指标。最后,根据本文的研究目的,结合本文的研究内容和成果,针对股票市场中的各个主体,分别提出了相应的对策建议。更多还原

【Abstract】 The existence of stock price manipulation has brought great harm to the stock market.On the one hand,its existence makes the stock price deviate from the real value of the stock,seriously harms the interests of ordinary investors,and weakens the enthusiasm of small and medium-sized investors to participate and investment confiden ce;On the other hand,the stock price man ipulation restricts the effective allocation of market resources,if it exists for a long time,it will gradually shrink the stock market,and ultimately will inevitably cause great losses to a country’s economy.In addition,since the establishment of China’s stock market,the case of stock price manipulation h,as been frequent,and most of them belong to the vicious manipulation behavior events.The relevant regulatory bodies are also constantly striving to improve the regulatory system,but with the strengthening of supervision,the traditional pattern of long-term manipulation behavior gradually disappeared,short-term manipulation behavior patterns continue to become the mainstream.The timing of the manipulation has changed from 1 years or more to the present 1-2 days,and the entire manipulation process is extremely short and ends in 1-2 days.Thus,it will pose a challenge and a threat to the ability of regulators to respond quickly.It is found that the existing index of stock price manipulation behavior identification and early warning index are basically financial transaction indicators(such as daily yield,volatility,turnover rate,etc.)and financial indicators.However,these indicators can only be calculated after the end of a trading day,so there is a serious lag in the existing discriminating methods.It simply can not immediately screen the stock market manipulation behavior,only applicable to the traditional long-term manipulation behavior,and does not apply to the current mainstream short-term manipulation behavior.In view of this,based on the existing research results,starting from the perspective of complex networks,this paper mainly studies from the following aspects:(1)Based on the China Securities Regulatory Commission found in 2015 in the Shanghai and Shenzhen A-share market trading stock manipulation cases as a sample,this paper used intraday pen tick transaction data,the stock buyers and sellers of the Commission declarations ID as a node,to the buyer and seller commissioned declarations whether the transaction for the connection to build a stock trading network.(2)Construct the model of stock trading manipulation behavior discrimination.Firstly,13 items,such as network density,number of nodes and number of connections,are screened by stepwise logistic regression method,which can describe the network parameters which are significantly different between the manipulated behavior stock and the non-manipulated behavior stock.Then,5 main explanatory factors are extracted by factor analysis method.Finally,based on the main factors extracted,the model of stock trading manipulation behavior discrimination is constructed.(3)To divide the community structure of the stock trading network constructed in the preceding article.In order to find the community structure hidden in the network topology,improve the accuracy of the early warning of stock price manipulation behavior.In this paper,from the perspective of Community structure,the Louvain algorithm,VOS clustering algorithm and GN algorithm are used to divide the built stock trading Network,and then compare which partitioning method works better.(4)The design of the early warning system of stock trading manipulation behavior is conceived.Aiming at the present situation that the cycle of manipulation behavior is getting shorter and the manipulation behavior is becoming more and more hidden,this paper selects the complex network parameters and the community structure parameters with high sensitivity as the early warning index,constructs the stock trading manipulation behavior Early Warning index by using the selected parameters,and divides the grade to the stock price manipulation behavior warning according to the comprehensive index score.It was eventually validated in one case.Through this article step by step research and analysis,in the end,mainly got the following three conclusions:(1)The model of stock trading manipulation behavior based on complex network parameter factors has a high accuracy rate.The empirical results show that the overall accuracy of the test in the sample of the discriminant model constructed in this paper is 86.60%,and the overall accuracy of the sample external test is 86.50%.The results of this paper can provide technical support for securities regulatory departments to effectively identify stock trading manipulation behavior,allocate regulatory resources rationally,and crack down on market manipulation.(2)In the division of the stock trading manipulation behavior Network,the algorithm suitable for dividing the network constructed in this paper is selected,and Vos clustering is the best compared with the division effects of VOS Clustering algorithm,Louvain algorithm and GN algorithm.In most cases,the division result of Louvain algorithm is better than the division result of GN algorithm.(3)The ring growth rate of the number of nodes,the ring growth rate of the number of elements,the growth rate of the maximum number of elements,the growth rate of the community number,the network density and the maximum group size(proportion)can be used as an early warning index of stock trading manipulation behavior.According to the research purpose of this paper,combined with the research contents and achievements of this paper,according to the various subjects in the stock market?the following countermeasures and suggestions are put forward respectively.更多还原