基于遗传神经网络上市公司股价向基本面回归的分析研究

发布时间：2018-05-19 06:15

本文选题：股价 + BP神经网络　；参考：《天津科技大学》2012年硕士论文

【摘要】：证券市场对国民经济的推动作用毋庸置疑,尤其是在经济全球化的形势下,证券市场能否健康发展直接影响国家的经济竞争力。在我国资本市场中存在一种观点,我国上市公司的股票业绩与基本面无关,价格失真。股票市场是一个极其复杂的非线性系统,然而神经网络具有很强的自学习、自适应和非线性逼近能力等特性。遗传算法根据适者生存的原理进行全局搜索,注重搜索未知区域,将遗传算法和神经网络算法结合起来,可充分发挥他们的优势。为此,本文采取遗传算法和神经网络技术对上市公司股票价格与基本面的关联性进行实证研究。本文详细分析了影响我国股票价格的诸多因素,从宏观经济走向、上市公司业绩、其他因素这三个影响我国股票价格的主要因素中确立了神经网络模型的输入参数,分别是每股收益、净资产收益率、流通股本、GDP (Gross Domestic Product)和CPI (Consumer Price Index);输出参数为股票价格。本文选择BP (Back-Propagation)神经网络建模。通过实验确定了模型的层数及神经元个数和权值矩阵。使用Levenberg-Marquardt算法对BP神经网络进行了改进,减少了BP神经网络的训练时间,提高了BP神经网络的收敛精度。通过遗传算法与神经网络结合,用遗传算法对BP神经网络的初始权值和阈值进行优化。构建了上市公司股价与基本面因素研究的模型。最后,使用MATLAB对模型进行仿真、训练、验证。结果表明：虽然很长一段时间股价脱离基本面运行,但是随着市场秩序的规范和股票市场的健康发展,上市公司的股价正逐步趋向价值回归。
[Abstract]:There is no doubt about the promotion of the securities market to the national economy , especially in the situation of economic globalization , the healthy development of the securities market directly affects the economic competitiveness of the country . In the capital market of our country , there is a view that the stock performance of the listed company is independent of the fundamental plane and the price is distorted .

The stock market is an extremely complex nonlinear system , however , the neural network has strong self - learning , self - adaptation and nonlinear approximation ability . Genetic algorithm is based on the principle of the survival of the author . It focuses on searching unknown regions , combining genetic algorithm and neural network algorithm to give full play to their advantages . To this end , the paper adopts genetic algorithm and neural network technology to study the relationship between stock price and basic surface .

This paper analyzes the factors that affect the stock price of our country , and establishes the input parameters of the neural network model from the three main factors that affect the stock price of our country from the macro - economic trend , the company ' s performance and other factors . The input parameters of the neural network model are the earnings per share , the yield of net assets , the current share , gross domestic product and CPI ( Consumer Price Index ) , and the output parameters are stock price .

By using Levenberg - Marquardt algorithm , the BP neural network is improved , the training time of BP neural network is reduced , the convergence accuracy of BP neural network is improved , and the initial weight and threshold of BP neural network are optimized by genetic algorithm .

Finally , using MATLAB to simulate , train and validate the model , the results show that the stock price of the listed company is gradually returning to value with the regulation of the market order and the healthy development of the stock market .
【学位授予单位】：天津科技大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP183;F275;F832.51

【参考文献】