导　　师： 何剑;吴列进

学科专业： 0251

授予学位： 硕士

作　　者： ;

机构地区： 广东财经大学

摘　　要： 股票属于一种高风险、高收益的投资，已成为现代生活中不可缺少的一部分，因此投资者们时刻关心股市，分析股市，研究价格趋势。股票市场中随机因素很多，导致股票价格波动表现出很强的不确定性，以致传统预测技术的效果并不理想。于是，建立一个合理的股票价格预测模型，具有重要的理论意义和实践价值。 本文通过深入分析股票原理，比较常见的股票预测方法，探讨BP神经网络在股票预测上的可行性。从原理上讲，神经网络是对股票交易的历史数据学习后实现对未来股票价格的预测。具体而言，BP网络通过对股票的历史数据的学习,不断地修正相应的权值、阀值，最终建立一个相对合理的模型。本文研究的是投机的超短线股票交易，与传统的投资理念有明显区别；预测的结果是来源于多次预测结果的分析，而非特指某一次预测。 本文提出了一种创新的研究思想——引入基于时间价值的动态权重误差函数，设计出一种基于时间价值的神经网络模型。本文认为：BP模型通过引入动态权重的方法，可以改变了原来BP模型单纯的拟合训练集数据，更灵活地择优而达到预测效果。据此，本文采用MATLAB软件选定医药行业的股票进行仿真实验。实证结果表明:与传统预测方法和BP神经网络相比，本文提出的模型准确率较高，明显降低预测误差，进一步提高了网络的泛化能力和模型预测精度，优化了股票价格预测效果。为了验证模型的经济和社会效益，本文设计了一种现实中可实现的模拟交易操作（T+0模型），验证了基于时间价值的BP模型的价值。 Stock is a kind of high-risk and high-return investment, and has become an indispensable partof modern life，so investors always care for stock market, analyze storks, and research on pricetrend. There are many uncertain factors, and therefore price volatility shows strong uncertainty, sotraditional prediction technology is unsatisfied. Establishing a logical model for stock priceforecasting has theoretical significance and applicable value. This paper analyzes the theory of stock, contrasts common stock forecasting methods, anddiscusses the feasibility of using BP neural network on stock price forecasting. In theory, neuralnetwork studies the historical data in order to forecast exchange price in future. Specifically, BPnetwork constantly revises its weights and valve values, through learning historical datum, in orderto establishing a relatively reasonable model finally. The paper is apparently different from ideas oftraditional investment, studying on ultra-short-term speculative stock trading; results are from theanalysis of many predictions, not just one in particular. The paper proposes a new way that is dynamic-weight error function based on time-value, anddesigns a neural network model based on time value by using dynamic weight. In the paper, BPmodel has changed the way of fitting training data by introducing the dynamic-weight method, andis designed more flexibility to meet the practical forecasts. This paper does simulation experimentusing MATLAB on stocks of the pharmaceutical industry. Empirical results show that, comparedwith traditional methods and BP neural network, the model proposed by this paper has higheraccuracy and lower error, and further improves the network’s generalization ability and predictionaccuracy of the model, and achieves better optimization of stock prediction. To verify theeffectiveness of the model on economic and social benefits, the paper designs the simulation thatcan be achieved in reality transactions /(T+0model/), and verifies its realistic prospect.

关 键 词： 时间价值 神经网络 股票价格预测 网络模型

分 类 号： [TP183 F830.91]

领　　域： [自动化与计算机技术] [自动化与计算机技术] [经济管理]