我国的农产品期货市场虽然经历了多年的发展，但与西方成熟的期货市场相比，还是不够健全。由于期货投资者以个人投资者为主，且受其市场信息获取不充分和投机心理重等原因，容易去跟风模仿别人的投资行为，使得整个交易市场出现“羊群行为”，造成市场的动荡。本文基于以上背景，采用复杂网络相关理论和多Agent技术，构建了一个基于复杂网络的农产品期市行为模型。全文主要工作如下： 首先，对真实的中国农产品期货市场的典型合约进行了实证分析。选取了大连商品交易所2008至2012年5年的大豆1期货合约价格进行了数据预处理，得到了收益率序列并对其进行了统计分析，得出收益率序列远远偏离正态分布，存在着尖峰厚尾的特征，并不服从随机游走规律。对序列进行的R/S分析结果表明，序列存在长记忆性。通过实证分析得到的这些典型特征也为随后要建立的复杂网络的农产品期市行为模型提供了模型有效性验证标准。 随后，对基于复杂网络的农产品期市行为模型进行了建模。主要包括三方面的工作：投资者Agent基本属性及投资行为建模，投资者人际网络关系建模，期货市场定价机制建模。在投资者Agent基本属性及投资行为建模中，把多Agent系统构造的思想和方法运用其中，首先把投资者分为大户、中户和散户三种不同的类型并赋予不同的财富量，同时，也确定了投资者所掌握信息、决策行为、对期货合约的报价以及订单量的表达式；在投资者人际关系网络建模方面，把投资者之间的联系分为邻居联系和远程联系两个方面，从而体现出了人际关系网络的小世界性。而在邻居联系方面，大户、中户和散户的邻居数目各不相同，从而体现了人际关系网络的无标度性。最后，在期货市场定价机制建模方面，本模型选取了真实期货市场中的一种定价方式——集合竞价，并予以实现。 最后，对第4章建立的模型进行了仿真实验。通过分析期货市场价值与价格的时间变化序列，以及买单卖单的变化情况，可以从直观上得出本文构造的期货市场行为模型并不符合随机游走规律，存在着羊群行为。为了使结论更具有说服力，又对本模型的价格收益率序列进行了定量的统计分析，分析结果表明，收益率序列不服从随机游走规律，具有尖峰厚尾特征，并且存在长记忆性。因此，可以认为本仿真模型再现了真实农产品期货市场的典型特征，本文所构造的仿真模型是有效的。最后，又对羊群行为与收益率之间的关系进行了研究，并认为两者之间不存在显著关系。; China’s produce futures market has been developing for years, but compared to the well-developed western market, it still has a long way to go. As the major investors in China’s produce futures market are individual investors, who cannot get full access to the market news and tend to mimic others’ behaviors, the market is liable to show herd behavior. Given this background, a behavioral model of produce futures market based on complex network (BMPFM-CN) is built under the guidance of related theories and techniques in complex network, multi-agent system and herd behavior. Firstly, an empirical analysis on the representative future contract – the future contract bean 1 from Dalian Commodity Exchange (DCE) is conducted. The price series from 2008 to 2012 is preprocessed and the earning rate series is gained. The statistical analysis shows the earning rate series is far from normal distribution, and has a significant feature of leptokurtosis and fat-tail. The R/S analysis result also indicates the series is with long-term memory. Those statistical analyses offer a validation standard for the model this study is about to build. Then, the BMPFM-CN is modeled. The modeling includes three tasks: 1) the modeling of basis attributes and investment behavior of investor agents: agents are divided into three types, the large, the middle and the small and diverse wealth are assigned to agents accordingly; expressions of agent’s information, investment behavior, quoted price and volume are settled; 2) the modeling of the agent’s interpersonal network: the network includes neighbourhood relationship and remote relationship. According to its type, each agent has different neighbours and probability of remote relationship. In this way, the whole network demonstrates small-world characteristic as well as scale-free one; 3) the modeling of pricing mechanism of BMPFM-CN: call auction is realized. Finally, simulation experiments are carried out. By analyzing the value and price series, and the buy and sell volume series, conclusion can be intuitively draw that the BMPFM-CN does not conform the law of random walk and the herd behavior does exist. To make the conclusion more convincing, a statistical analysis of the earning rate series is also conducted. The result shows that the series has a feature of leptokurtosis and fat-tail, and also the long-term memory. Therefore, it can be drawn that the BMPFM-CN can reproduce the features shown in the real futures marke...
修改评论