基于主成分分析和BP神经网络算法的综合选股实证研究

发布时间：2018-03-30 11:25

本文选题：股价预测　切入点：投资指导　出处：《暨南大学》2017年硕士论文

【摘要】：随着金融业和计算机技术的快速发展,为获取超额或稳定投资收益的量化选股研究在我国已经逐渐兴起。因此基于机器学习算法构建量化模型进行股价预测和投资指导十分有必要,但股票市场是一个复杂的非线性系统,利用传统算法构建模型作为股价预测和投资指导的工具,具有很大的局限性,且传统量化模型偏单一。因此本文将基于主成分分析法进行上市企业财务状况评价研究、基于主成分分析法与BP神经网络的结合方法进行股价预测研究和传统的选股分析方法组合构建多样化的量化选股模型。因为近年来神经网络对于处理非线性问题展现出了独特的优势,神经网络通过对股票历史数据的学习,将股价变化的规律存储在神经元的权值中,通过运用训练好的网络来预测未来股价。本文的主要工作具体为:首先通过运用主成分分析法对深圳A股市场上广东地区的上市公司主要财务指标数据进行分析得到了评价指标综合得分,依据综合得分可以直接且全面的了解上市公司的财务状况,实证研究表明该模型评价方法的有效性。其次本文将财务指标数据和交易指标数据同时作为股价预测研究的输入变量,实证研究发现基于主成分分析法和BP神经网络的结合方法进行股价预测能够有效且准确度较高的评判股票下一年同季度的涨跌情况,我们采用了三种BP神经网络算法进行实验,其中基于贝叶斯正则化算法的BP神经网络构建的模型预测准确度最高且模型训练过程中能够有效避免过拟合现象,预测值与实际值误差较小。最后本文的综合选股方案为:先选取基于主成分分析法进行上市企业财务状况评价研究所推荐的前25%的股票,然后再选取基于主成分分析法与BP神经网络的结合方法进行股价预测研究所推荐的前25%的股票,最后结合传统的选股分析方法综合分析选取上市企业财务状况评价研究和股价预测研究两者共同推荐的前25%的股票中的小盘股作为下一年年中最优的投资股票。
[Abstract]:With the rapid development of finance and computer technology, In order to obtain excess or stable investment income, the research of quantitative stock selection has gradually risen in China, so it is necessary to build a quantitative model based on machine learning algorithm for stock price prediction and investment guidance. But the stock market is a complex nonlinear system. It is very limited to use the traditional algorithm to construct the model as the tool of stock price prediction and investment guidance. Therefore, based on the principal component analysis method, the financial status evaluation of listed enterprises will be studied in this paper. Based on the combination of principal component analysis (PCA) and BP neural network (BP neural network), stock price prediction and traditional stock selection analysis are combined to construct a diversified quantitative stock selection model. The neural network has been used to deal with nonlinear problems in recent years. Showing unique advantages, The neural network stores the law of stock price change in the weight of neuron by learning the historical data of stock. The main work of this paper is to analyze the main financial index data of listed companies in Shenzhen A share market by using principal component analysis method. The comprehensive score of evaluation index is obtained, According to the comprehensive score, we can directly and comprehensively understand the financial situation of listed companies, The empirical study shows the effectiveness of the evaluation method. Secondly, the financial index data and the transaction index data are taken as the input variables of the stock price forecasting research. The empirical study shows that the combination of principal component analysis and BP neural network can effectively and accurately judge the stock price fluctuation in the same quarter of next year. We use three BP neural network algorithms to carry out experiments. The BP neural network based on Bayesian regularization algorithm has the highest prediction accuracy and can avoid overfitting effectively in the process of model training. The error between the forecast value and the actual value is small. Finally, the comprehensive stock selection scheme of this paper is as follows: firstly, the first 25% stocks recommended by the Institute of Financial Evaluation of listed Enterprises based on Principal component Analysis (PCA) are selected. Then select the method based on the combination of principal component analysis and BP neural network to carry out the first 25% stocks recommended by the Research Institute for Stock Price Prediction. Finally, combined with the traditional stock selection analysis method, the paper selects the small-cap stocks of the top 25% stocks recommended by both of them as the best investment stock in the middle of next year.
【学位授予单位】：暨南大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP183;F832.51

【参考文献】