基于人类主体实验的学习模型校验

发布时间：2018-04-16 21:19

本文选题：人类主体实验 + Roth-Erve学习模型　；参考：《天津大学》2014年硕士论文

【摘要】：计算实验金融学研究中的构建人工股票市场,核心问题是主体的学习机制的设计,学习机制是主体理性不同的体现。随着大量学习模型的理论研究的涌现,采用实际的、高质量的实验数据来对模型进行校验研究的重要性也日益凸显。本文实际采用人类主体实验的方法获得投资者学习过程的股价序列,来校验计算实验金融里Roth-Erve学习模型中的遗忘参数和类比参数。通过在一定区间内改变学习模型参数的大小,得到拟合性相对高、适应性较强的投资者学习过程的模型参数。本文首先从计算实验金融研究的整体研究高度出发,详细探讨了Roth-Erve学习模型在此领域研究的起源、发展和重要应用意义,以及应运而生的学习模型的校验研究。并通过文献的梳理和对问题的理解,提出了采用适应我国投资者学习心理和过程的人类主体实验对学习模型校验的观点。因为校验学习模型的本质是从心理学角度出发的,单纯的将西方的理论套用在我国投资者上,其科学性和可行性是值得质疑的。通过基于实验经济学软件z-Tree,结合Hommoes提出的“学习并预测的主体实验”(LtFEs)的实验原理,设计了符合本国投资者心理特征的人类主体实验,实验后获得了股价序列等相关数据。同时,也基于MATLAB软件设计了采用Roth-Erve学习模型的计算金融实验,模拟投资者的投资决策过程,同样收集股价序列等数据。通过调整学习模型的参数值,来控制学习模型的拟合效果,进而探析如何设置学习模型参数能够更好的模拟我国投资者的心理特征和学习过程。实验结果表明,本文设计的人类主体实验得到具有收敛和阻尼振动特征的两类股价序列。当价格序列收敛时,说明本实验所构造的异质投资者条件和自适应的学习过程是有效的,能够达到类似理性预期均衡的状态;通过计算金融实验,用Roth-Erve学习模型对两类股价序列进行模拟,能够找到拟合性高的解,并结合四个选项的实际意义,分析了学习过程和股价序列特征的理论原因;最后,在给定的参数区间内,通过改变参数而进行拟合性高低的比较,可以找到相对本国投资者适应性更高的参数值。这为计算实验金融的应用,提供了更适应本国投资者的学习模型。
[Abstract]:The key problem in the construction of artificial stock market in the research of computational experimental finance is the design of the learning mechanism of the subject, which is the embodiment of the different subjective rationality.With the emergence of a large number of theoretical studies on learning models, the importance of using practical, high-quality experimental data to verify the models is becoming increasingly prominent.In this paper, the stock price sequence of investor learning process is obtained by using the method of human subject experiment in order to verify and calculate the forgetting parameters and analogical parameters in the Roth-Erve learning model in experimental finance.By changing the parameters of the learning model in a certain interval, the model parameters of the investor learning process with relatively high fit ability and strong adaptability are obtained.In this paper, the origin, development and important application of Roth-Erve learning model in this field are discussed in detail from the perspective of the overall research on computational experimental finance.By combing the literature and understanding of the problem, this paper puts forward the viewpoint of checking the learning model by using the human subject experiment which adapts to the learning psychology and process of the investors in our country.Because the essence of the calibration learning model is from the angle of psychology, it is doubtful that the western theory is applied to the investors in our country.Based on the experimental economics software z-Treeand combined with the experimental principle of "Learning and predicting the subject experiment" proposed by Hommoes, this paper designs a human subject experiment which accords with the psychological characteristics of domestic investors. After the experiment, the stock price sequence and other relevant data are obtained.At the same time, based on the MATLAB software, we also designed a computational financial experiment using Roth-Erve learning model to simulate the investors' investment decision-making process, and also collect the stock price sequence and other data.By adjusting the parameters of the learning model, the fitting effect of the learning model is controlled, and how to set the parameters of the learning model can better simulate the psychological characteristics and learning process of the investors in China.The experimental results show that two kinds of stock price sequences with the characteristics of convergence and damping vibration are obtained in the human subject experiment designed in this paper.When the price sequence converges, it shows that the heterogeneity investor condition and adaptive learning process constructed in this experiment are effective and can achieve a state similar to rational expectation equilibrium.The Roth-Erve learning model is used to simulate two kinds of stock price sequences, which can find a good fitting solution. Combined with the practical significance of the four options, the theoretical reasons of the learning process and the characteristics of the stock price sequence are analyzed. Finally, in the given parameter range, the theoretical reasons for the characteristics of the learning process and the stock price sequence are analyzed.By changing the parameters and comparing the fitness, we can find the parameter value with higher adaptability than the domestic investors.This provides a more suitable learning model for domestic investors for the application of computational experimental finance.
【学位授予单位】：天津大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：F830.91

【参考文献】