基于复杂网络的证券市场传闻扩散与羊群行为演化研究

发布时间：2018-06-20 19:54

本文选题：市场波动 + 复杂网络　；参考：《南京航空航天大学》2012年博士论文

【摘要】：证券市场无疑是一个复杂系统，至少包括三个基本构成成份：交易者（Actors）、股票（Stocks），以及影响投资者交易行为和股价波动的信息（Event或Rumors）。这三者之间不是相互孤立的，而是相互作用、相互影响，使证券市场表现出复杂的运行行为。交易者、股票以及市场信息之间存在着错综复杂的联系，从而构成了各种各样的网络形式。本文构建了四种类型的网络分别研究传闻在市场中的扩散规律和羊群行为的生成机理：交易者网络、股票相似性网络、交易者-股票的二分网络以及交易行为网络。这些网络都是复杂的，因此本文采用复杂网络的理论与方法，不但分析网络的拓扑结构，而且分析网络上的动态演化行为。论文首先采用系统动力学模型分析了证券市场波动性影响因素，以及证券市场波动性与传闻扩散、羊群行为的关系；然后，构建了基于交易行为的证券市场波动模型，并分析了卖空限制交易和事件驱动机制对于市场波动性的影响；其次，通过构建交易者网络分析市场传闻的博弈扩散过程与深度，通过构建股票相似性网络分析传闻在市场中的扩散分布；再次，通过构建交易者-股票的二分网络模型，分别采用理论推导和仿真实验的方法分析了市场波动性以及羊群行为；通过构建以股票为节点的交易行为网络，分别采用理论推导和仿真实验的方法分析了羊群行为的生成机制；最后，给出了研究结论与进一步研究展望。全文共分为5个部分。第一部分给出了论文的研究背景、研究问题、以及研究的目的和意义，对证券市场的信息不对称性、投资者羊群行为、以及复杂网络的理论与方法等的研究现状进行了综述，确定了本文的研究方向与研究重点；最后，给出了本文的研究方法、技术路线、体系结构和创新点。第二部分采用系统动力学方法分析了证券市场波动性的影响因素。首先，对证券市场波动性的影响因素进行了分类，共分为五种类型的影响因素：宏观基本面、中观基本面、微观基本面、技术面和隐形面；然后，构建了证券市场波动性影响因子系统的动力学模型，通过Vensim仿真发现，证券市场波动性直接受到技术面和隐形面因素的影响，与其同步变化，而三类基本面因素则稍微滞后于波动性的变化；最后，构建了证券市场波动性与传闻扩散、羊群行为的系统动力学概念模型，通过正负反馈回路的分析发现，传闻扩散是各种因素影响波动性的集中体现与重要方式，羊群行为则是市场波动性在交易者群体上涌现出的一种集体行为方式。第三部分从限制卖空交易和事件驱动两个角度分析了证券市场的波动性。对于限制卖空交易的研究发现，卖空限制降低了了市场效率，增强了波动性；卖空限制强化了单边市场，降低了市场流动性；卖空限制阻碍了价格发现，助长了羊群效应。对于事件驱动机制的研究发现，股票价格波动率的变化与作用强度成正比，与当前价格波动率成正比，与未来增值空间成正比。在一定时间内股票价格走势曲线在事件的驱动下近似服从于S-曲线。在投资资金约束下，股票价格波动率变化曲线从单调递增的S-曲线变为具有最大值的单峰曲线；在市场走势与事件驱动方向相同的情况下，市场指数将放大事件驱动的影响，加速股票价格波动率的上升；在市场走势与事件驱动方向相反的情况下，市场指数将打压事件驱动的影响，使股票价格波动率变化曲线成为具有最大值的单峰曲线。最后，本文提出了一个“矩形窗口”的方法指导投资者判断潜在事件对股票价格波动率的影响。第四部分从交易者网络博弈和金融相似性网络等两个角度分析了传闻在市场中的扩散规律及其影响。对于传闻的博弈扩散研究发现，在Nash均衡点，知情者采取扩散策略的概率与不知情者的接受成本和拒绝代价之差成正比；不知情者采取接受策略的概率不但与知情者的封锁成本成正比，而且与网络中不知情者的平均度成正比；从转移矩阵可知，市场传闻扩散的马尔科夫链是一个吸收链，从而市场传闻的扩散状态将最终变为封锁或拒绝状态，并且市场传闻扩散的平均步数与网络的平均邻接不知情者的数目成正比。基于金融相似性网络的传闻扩散研究发现，由股票时间序列的相关系数矩阵，，得到了两类网络：完全加权相似性网络和正负加权相似性网络。完全加权相似性网络具有全连通的拓扑结构，是一个规则网络。其节点连通度分布属于单点分布，而节点连通度相关的条件概率等价于从所有边中随机选取一条的概率；正负加权相似性网络是无标度网络，其拓扑结构是不均匀的。将传闻扩散模型应用于完全加权相似性网络，其结果恰好与经典的传染病扩散模型SI相同；将传闻扩散模型应用于正负相似性网络，其结果与网络的复杂性密切相关。第五部分从股票-投资者的二分网络和交易行为网络研究了羊群行为的生成机理。在二分网络上研究发现，随着交易者之间学习概率的不同，羊群行为表现为交易者持股情况的三种分布：当交易者之间的学习概率p p很大，交易者持股近似服从脉冲分布，股票市场表现出很强的羊群行为；当交易者之间的学习概率p p较小，交易者的持股近似服从二项分布，股票市场表现出非常弱的羊群行为；当交易者之间的学习概率p p处于一个适当的范围时，交易者的持股服从不同的具有指数截断的幂律分布，股票市场表现出不同程度的羊群行为。在交易者网络上研究发现，当交易者之间的学习模仿概率满足01时，市场中的羊群行为服从幂指数为3的幂律分布；当交易者之间的学习模仿概率满足0时，市场中交易行为分布服从指数分布，不存在羊群行为。最后，总结了全文的研究工作，并就进一步研究的方向进行了简要的讨论。
[Abstract]:The securities market is undoubtedly a complex system, at least three basic components: the Actors, the Stocks, and the information (Event or Rumors) affecting the trading behavior and the volatility of the stock. These three are not isolated from each other, but interact and interact with each other to make the securities market a complex running line. Yes.
There are intricate links between traders, stock and market information, which constitute a variety of network forms. This paper constructs four types of networks to study the spread of rumours in the market and the generation mechanism of herd behavior: the trader network, the share ticket similarity network, the trader and the stock's two point network. And the transaction behavior of the network. These networks are complex, so this paper uses the theory and method of complex network, not only the analysis of the topology of the network, and the analysis of dynamic evolution behavior on the network.
Firstly, the paper analyzes the influence factors of volatility in securities market, and the relationship between the volatility of securities market and the rumor diffusion and the herding behavior. Then, the volatility model of securities market based on transaction behavior is constructed, and the effect of short selling limited transaction and part driving mechanism on the market volatility is analyzed. Second, we analyze the diffusion process and depth of the game by constructing the trader network and analyze the spread distribution in the market by constructing the stock similarity network. Thirdly, the market volatility and the herd line are analyzed by the method of theoretical deduction and simulation experiment by constructing the two point network model of the trader and stock. For; by building a stock for the network transaction behavior of nodes, methods, theoretical derivation and simulation analysis of the formation mechanism of the herd behavior; finally, the prospect of further research and conclusions are given.
The full text is divided into 5 parts. The first part gives the research background, the research problem, the purpose and the significance of the research, the information asymmetry of the securities market, the herd behavior of the investor, the theory and method of the complex network, and determine the research direction and the key point of this paper. Finally, The research method, technical route, system structure and innovation point are given.
The second part analyzes the influence factors of the volatility of the securities market by using the system dynamics method. Firstly, it classifies the factors affecting the volatility of the securities market, which are divided into five types of influencing factors: macro fundamentals, meso fundamentals, micro fundamentals, technical and invisible surfaces; then, the volatility shadow of the securities market is constructed. The dynamic model of the sound factor system, through the Vensim simulation, shows that the volatility of the securities market is directly influenced by the technical and invisible factors, and the three basic factors are slightly lagging behind the volatility. Finally, the system dynamics of the volatility of the stock market and the rumor diffusion and the herding behavior are constructed. Read model, through the analysis of positive and negative feedback loop that rumors diffusion is the various factors affecting embodied volatility and an important way of herd behavior is a kind of collective behavior of the emergence of market volatility in the group of traders.
The third part analyzes the volatility of the stock market from two aspects of limited short selling and event driven. For the study of short selling, the short selling limit reduces the market efficiency and increases the volatility; short selling limit strengthens the unilateral market and reduces the liquidity of the market; short selling limit hinders the price discovery and encourages sheep. Group effect. The study of event driven mechanism found that the change of stock price volatility is proportional to the intensity of action, proportional to the current price volatility, and proportional to the future value added space. In a certain time, the trend curve of stock price is approximately subject to the S- curve driven by events. Under the constraint of investment funds, the stock price wave When the market trend is the same as the event driven direction, the market index will enlarge the impact of the event driven and accelerate the rise of the volatility of the stock price. In the case of the opposite trend of the market trend and the event driving direction, the market index will press things under the situation of the same market trend and the event driven direction. The influence of driving, the volatility of the stock price curve has become a unimodal curve maximum. Finally, this paper puts forward a method of a rectangular window "to guide the investors to judge the impact of potential events on the rate of stock price volatility.
The fourth part analyzes the spread and influence of hearsay in the market from two angles, such as the network game of traders and the network of financial similarity. In the study of the rumor game diffusion, it is found that in the Nash equilibrium point, the probability of adopting the diffusion strategy by the insider is proportional to the difference between the acceptance of the unaware person and the rejection price; The probability of taking the acceptance strategy is not only proportional to the blocking cost of the insider, but also proportional to the average degree of the unknown person in the network. From the transfer matrix, the Markoff chain of the market rumor diffusion is an absorption chain, and the diffusion state of the market hearsay will eventually become blockade or refusal, and the market rumor is spread. The mean step is proportional to the number of the average neighbor of the network. The study of the rumor diffusion based on the financial similarity network finds that the two types of networks are obtained from the correlation coefficient matrix of the stock time series: the fully weighted similarity network and the positive and negative weighted similarity network. Structure is a regular network. The node connectivity distribution belongs to the single point distribution, and the conditional probability of the node connectivity is equivalent to the probability of random selection from all sides; the positive and negative weighted similarity network is a scale-free network and its topology is not uniform. The spread model is applied to the fully weighted similarity network. As a result, coincided with the classical infectious disease diffusion model SI will be the same; rumor spreading model is applied to the positive and negative similarity network, which results in complexity and network are closely related.
The fifth part studies the generation mechanism of herd behavior from the stock investor's two division network and the transaction behavior network. On the two point network, it is found that with the difference of the learning probability between traders, the herd behavior is the three distribution of the trader's Stock Ownership: the learning probability of the trader is p p, and the trader holds the stock close. Similar to the pulse distribution, the stock market shows a strong herd behavior; when the learning probability of p p between traders is smaller, the stock market shares approximately two distribution, and the stock market shows very weak herding behavior; when the learning probability p p between traders is in a proper range, the stock ownership of the traders is different from that of the traders. A power law distribution with exponential truncation shows a different degree of herding behavior in the stock market. On the trader's network, it is found that when the probability of learning imitation between the traders satisfies 01, the herd behavior in the market obeys a power law distribution of 3. When the learning imitation probability of the trader meets 0, the market transaction is in the market. The behavior distribution obeys exponential distribution, and there is no herding behavior.
Finally, the research work is summarized, and the direction of further research is briefly discussed.
【学位授予单位】：南京航空航天大学
【学位级别】：博士
【学位授予年份】：2012
【分类号】：F224;F830.91

【引证文献】