基于多元竞争性做市商的人工金融市场仿真实验研究

发布时间：2018-04-04 10:14

本文选题：Agent　切入点：计算经济学　出处：《华中科技大学》2013年硕士论文

【摘要】：近年来，我国OTC市场发展迅猛。但是OTC市场具有上市证券具有价格低、流动性差、风险大等特点，单一的竞价制度并不足以解决流动性差的问题，而且当有大宗交易发生时，极易导致竞价市场的价格剧烈震动，给交易者带来很高的风险，在这样的背景下，我国金融市场是否该引入做市商制度的讨论引起了广泛关注。金融市场中的各种复杂经济现象是特定市场结构下市场参与主体通过微观层面的相互作用在宏观层次的系统涌现，基于Agent建模的计算经济学为复杂金融系统的研究提供了一个统一、开放的研究方法。本文根据复杂适应理论和基于Agent建模的计算经济学原理，通过引入多元做市商报价制度，利用Agent建模技术构建了一个基于多元竞争性做市商的人工金融市场仿真框架，研究这种特定的市场结构下市场的涌现特征。首先，在对真实市场中做市商、投资者、市场环境这3大组成部分进行合理抽象的基础上建立了做市商、投资者决策行为的数学模型和市场环境的结构。然后在ANYLOGIC仿真软件上开发做市商Agent、投资者Agent、市场环境Agent类，通过定义多种方法模拟各类Agent的行为实现了基于多元竞争性做市商的人工金融市场仿真框架，为了对仿真数据进行统计分析，本模型还开发了一个将数据导入EXCEL存储的类。最后，在此仿真框架下，进行了仿真模型合理性检验实验和做市商报价特性研究实验。模型合理性验证实验结果显示：日收益率时间序列复现了真实金融市场中典型的尖峰厚尾特征；日收盘成交价紧随真实价值上下稳定波动，符合做市商能平抑股价波动的特点；市场价差小于每个做市商单独做市时的市场价差，符合多元竞争性做市商能提高市场流动性的特点。上述这些与现实相吻合的特征表明本文所建立的模型是合理的。做市商报价特性研究实验结果表明：两个做市商在每个交易日的报价均能收敛于真实报价，其中具有动态学习能力的做市商的收敛偏差较采用静态预测方法的做市商小，，但其收敛速度却慢一些。做市商报价对真实价值的收敛性证明了本文所设计的做市商报价决策模型是有效的。
[Abstract]:In recent years, China's OTC market is developing rapidly.However, the OTC market has the characteristics of low price, poor liquidity, high risk and so on. A single bidding system is not enough to solve the problem of poor liquidity.It is easy to cause price shock in bidding market and bring high risk to traders. Under this background, the discussion on whether to introduce market maker system into financial market in our country has aroused widespread concern.Various complex economic phenomena in financial markets are the emergence of the market participants at the macro level through the micro level interaction under the specific market structure.Computational economics based on Agent modeling provides a unified and open approach for the study of complex financial systems.According to the theory of complex adaptation and the principle of computational economics based on Agent modeling, this paper constructs an artificial financial market simulation framework based on multiple competitive market makers by introducing multiple market-makers quotation system and using Agent modeling technology.This paper studies the characteristics of market emergence under this particular market structure.Firstly, the mathematical model of market maker and investor's decision behavior and the structure of market environment are established on the basis of reasonable abstraction of market maker, investor and market environment.Then the Agent classes of market maker agent, investor agent and market environment Agent are developed on the ANYLOGIC simulation software. The artificial financial market simulation framework based on multiple competitive market makers is realized by defining a variety of methods to simulate the behavior of all kinds of Agent.In order to analyze the simulation data, a class to import the data into EXCEL storage is developed.Finally, the rationality test experiment of simulation model and the research experiment of market quotation characteristic are carried out under this simulation framework.The results of the model rationality test show that: the time series of daily yield reproduces the typical sharp and thick tail characteristics of real financial market, the daily closing trading price follows the stable fluctuation of real value, which accords with the characteristics of market makers' ability to stabilize the volatility of stock price;The market price difference is smaller than the market price difference when each market maker makes a single market, which accords with the characteristics that multiple competitive market makers can improve market liquidity.These characteristics which are consistent with reality show that the model established in this paper is reasonable.The experimental results show that the price of two market makers can converge to the real price in each trading day, and the convergence deviation of the market maker with dynamic learning ability is smaller than that of the market maker with static forecasting method.But the convergence rate is slower.The convergence of market maker quotation to real value proves that the decision model designed in this paper is effective.
【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：F832.5;F224

【参考文献】