基于网络的金融数据分析与挖掘

发布时间：2018-01-06 00:31

本文关键词：基于网络的金融数据分析与挖掘　出处：《复旦大学》2013年博士论文　论文类型：学位论文

【摘要】：经济物理学是一个利用物理学的理论方法分析解决经济和金融问题的跨学科研究领域。它开启了物理和计算机研究者们对经济过程和金融市场问题的研究热潮。复杂网络是一个研究真实复杂系统的强大而有效的工具。但是,利用复杂网络方法研究金融市场还处于起步阶段,没有深入的实证研究。本文从复杂网络的理论框架对期货市场的网络统计特征、演化建模、群体行为和攻击弹性,以及从数据挖掘的角度对异常交易行为检测和价格传导机制进行深入地分析研究,希望能够帮助人们理解金融市场,认识其运行机制和规律,以及给监管者在预防极端市场和降低系统性风险的分析决策上提供有价值的思路。1.介绍期货交易网络的构建方法,全面分析期货交易网络的拓扑特征。期货交易网络是基于成交结果数据集构建的,节点代表交易者,边表示交易者之间的买卖交易关系。实证分析的结果表明期货交易网络具有这些特性：无标度行为、奇偶度分叉现象、小世界效应、层次组织、幂律介数分布、异配连接模式以及平均路径长度和网络直径随着网络生长而收缩的特性。对于期货交易网络每一个特性,我们均从期货市场交易业务和运行机制上做了解释和说明。实证分析的结果能够帮助市场参与者们和研究者们认识和理解金融市场交易业务的内在性质,也能够给金融市场监管者们在制定风险管控政策和具体措施时提供有价值的参考。2.分析期货市场超级操盘手的集体行为特性,并探究对其目标攻击后的网络弹性。我们对网络拓扑演化、富人俱乐部系数和目标攻击的场景进行研究分析,研究结果首先显示超级操盘手们特有的行为特征：交易次数多,交易时间长,交易品种繁杂,长时间保持活跃,拥有更多的机会进行再次交易；同时,他们在市场交易行为上显示了领导者的地位；其次,期货交易网络展示了明显的富人俱乐部结构,超级操盘手们互相紧密连接在一起；最后,期货交易网络在静态蓄意攻击下显示了很好的稳定性和健壮性,但是在每日交易网络的目标攻击弹性上,期货交易网络的结构演化表现各异：铜期货网络受到显著的冲击,天然橡胶期货网络没有影响,全市场网络有轻微的影响。这些分析结果可以帮助制定有效的监管方法,减弱金融市场系统性风险。3.跟踪期货交易网络的演化过程,揭示网络连接的微观动力学,并创建网络模型。期货交易网络经历了开始阶段的波动后,逐渐达到稳定状态,最大连通分量自始至终统治着网络的发展路线。在演化的过程中,期货交易网络一直展示了度和强度分布的无标度行为、小世界效应和异配连接模式。在微观层面上,我们揭示了网络连接的动力学,它作为网络演化的原始驱动力,并深入解释奇偶分叉现象。根据本质上越活跃的投资者会做越多买卖交易这个事实,以及网络连接动力学的分析结果,我们提出了一个活跃度模型,它能够成功地复制实证观察的结果。这些研究发现和建模思想可以帮助理解金融市场发展轨迹和运行机制。4.提出了一种金融市场里关联交易行为的检测方法。为了捕捉金融市场里这种新近出现的且呈上升发展趋势的异常交易行为,我们提出一个在期货合约上检测潜在关联交易组的方法：计算任何两个符合条件带符号委托量的统一聚集时间序列的相关系数,形成一个相关系数矩阵,构建相应的有权图；将多个每日有权图中的连通分量合并为综合有权图,多次出现的连通分量则认定为是一个潜在的关联交易组。我们在真实的委托记录集上进行检测实验,实验的结果显示我们的方法能够有效地检测期货市场上可疑的关联交易组,经过资深的业务专家验证,这个结果是可信的。5.提出基于延时相关性的价格传导关系图和趋势影响的方法,研究分析三家世界主要铜期货市场(LME、上期所和COMEX)的价格传导和互动关系。根据三家交易所每日两两价格序列间的延时相关性计算结果,我们建立价格传导关系图。对价格传导关系图的分析结果表明：LME领导着全球铜期货市场的价格,COMEX紧紧追随着LME,上期所价格受到LME强势的引导和影响。但是上期所价格和LME之间存在着明显的偏差。我们利用1分钟高频价格数据计算延时相关性,没有发现LME和上期所之间存在价格上的领导跟随关系。于是,我们利用趋势影响模型,测量上期所和LME在分钟级涨跌趋势上先背离后趋同的程度,结果表明上期所和LME在日内价格运行上存在明显的互动现象,但是LME对上期所的影响稍大。从三家交易所隔夜价格变化和交易时间的分析结果来看,上期所的交易时间太短,严重削弱其在全球市场价格发现中的地位。
[Abstract]:Physics is an economic analysis method by using the theory of physics interdisciplinary research field to solve economic and financial problems. It opens the study of physics and computer researchers on the economic process and financial market research boom. Complex network is a complex system of real powerful and effective tool. However, the research on financial market the complex network method is still in the initial stage, no in-depth empirical research. This paper from the theoretical framework of the complex network of futures market network statistical characteristics, evolutionary modeling, group behavior and attack resilience, in-depth analysis and research from the perspective of data mining on the abnormal trading behavior detection and price transmission mechanism, hoping to help people understand the financial market, understand its operation mechanism and rules, and to regulators in the extreme market analysis to prevent and reduce systemic risk decision Policy provides the ideas of value.1. introduced the construction method of futures trading network, a comprehensive analysis of the topological characteristics of futures trading network. Futures trading network is constructed based on the transaction results in data sets, the nodes represent traders, traders and edges represent the relationship between sale and purchase transaction. The empirical results show that futures trading network has these characteristics: the scaling behavior of parity, bifurcation, small world effect, hierarchical organization, power-law distribution characteristics of betweenness, diameter shrinkage with the network growth with different connection mode and the average path length and network. For the futures trading of every network characteristics, we are from the futures market trading business and operating mechanism is explained and description. The results of empirical analysis can help market participants and researchers to know and understand the financial market transactions of inner nature, can also give financial city The collective behavior of field regulators provide valuable reference in the development of.2. risk management policies and specific measures of the futures market super trader, and explore the network attacks on its elastic targets after the evolution of network topology. We, the rich club coefficient and target attack scenarios were analyzed, the results show the first act the unique characteristics of super traders: many transactions, transaction time, transaction types are complicated, long time to stay active, have more opportunities to trade again; at the same time, they are in the market for display on the leader's position; secondly, futures trading network shows the rich club structure, super trader are closely connected together; finally, futures trading networks show good stability and robustness in static intentional attack, but in the daily transaction Target attack on the elastic network, the performance of different structure evolution of futures trading network: Copper Futures Network is a significant impact, did not affect the natural rubber futures network, has a slight impact on the whole market network. These results can help to develop effective monitoring methods, reduce systemic risk in financial markets.3. tracking futures trading network evolution process and reveal the microscopic dynamics of network connection, and create a network model. The futures trading network has experienced the beginning stages of the fluctuation, gradually reached a steady state, the development line of the largest connected component from first to last ruled network. In the process of evolution, the futures trading network has been demonstrated and scale-free behavior of intensity distribution, small world effect and with different connection mode. At the micro level, we reveal the dynamics of network connection, it is the original driving force of network evolution, And further explanation of parity bifurcation. Based essentially on the more active investors will do this fact more transactions, and network connection dynamics analysis results, we propose an active model, it can successfully replicate the empirical observation results. The study found and modeling idea can help to understand the financial market development and the operation mechanism of.4. proposed a detection method of financial market transactions. In order to capture the abnormal trading behavior of financial markets in the emerging and developing tendency, we propose a detection method of potential transactions group in futures contracts on the calculation of any of the two eligible unified signed entrusted amount aggregation time series correlation coefficient, a correlation coefficient matrix, construct the corresponding weighted graph; multiple daily weighted graph in connected component The amount with the comprehensive weighted graph, connected component appeared several times is identified as a potential transaction group. We in the real records of authorization sets were detection experiments, experimental results show that our method can effectively detect the related transactions on the futures market group suspected, after verification of senior business experts. This result is credible.5. method is proposed to influence the price transmission relationship graph and trend delay based on correlation, analysis of three of the world's major copper futures market (LME, SHFE and COMEX) price conduction and interactive relationship. According to the delay correlation three exchanges daily 22 price series between the calculation results, we establish the price conduction the diagram analysis results of price conduction diagram shows that the LME led global futures prices, COMEX closely with LME, the price is LME strong The guide and influence. But between the price and LME exist obvious deviation. We use 1 minute high-frequency price calculation delay correlation data, did not find the existence of the price leader to follow the relationship between LME and the previous period. Then, we use the influence trend model, measuring the LME first and after departure from the degree of convergence in the minute change trend. The results show that the LME and interactionwhich is obvious in price operation days, but the effect of LME on the slightly larger. From the analysis results of overnight price changes and trading time three exchanges, the transaction time is too short, seriously weakened its position in the global market the price discovery.

【学位授予单位】：复旦大学
【学位级别】：博士
【学位授予年份】：2013
【分类号】：F832.5;TP311.13

【参考文献】