多分辨小波神经网络在股票市场预测中的应用

发布时间：2018-01-18 10:20

本文关键词：多分辨小波神经网络在股票市场预测中的应用　出处：《华中师范大学》2016年硕士论文　论文类型：学位论文

【摘要】：股票市场是国家宏观调控、企业直接融资的重要领域,其收益和风险成正比。由于股票价格随机游走的特征,且受国内外经济政治变化、股民心理特征等众多因素影响,很难进行科学评测。股票价格影响因素复杂,呈高度非线性,利用传统的统计模型对有高噪声、非线性等众多特征的股价进行预测难以取得理想的效果。因此,建立一个准确度高、精简实用的评价模型对于证券市场投资及国家宏观调控都具有重大的现实意义。目前对于股票价格预测,绝大多数投资者采用的依据股票趋势、图形形状、人气指标进行技术分析,由于分析方法众多,缺乏科学系统的理论支持,且各指标的独立性较强,因此预测准确性不高。人工智能是一门集大成的科学,涵盖了计算机、心理学、图像处理等知识,在近年来取得了突破性进展和广泛的应用,神经网络是人工智能的一个分支,小波神经网络是基于传统网络之上,引入小波变换对其进行改造,既有神经网络的非线性逼近、自组织学习性、结构简单等特点,同时兼具小波分析的黑箱辨识能力,能极大增强预测的效果。2015年股市经历了杠杆疯牛,千股涨停、跌停、停牌,政府仓促出台救市措施,注定这是不平凡的一年。本文首先介绍了2015年我国股灾发生的详细经过,从宏观经济角度分析股灾的成因,继而介绍了相关的知识背景,包括股票市场的基础知识,现阶段股价的预测方法,神经网络和小波分析相关概念,小波神经网络的基本特征以及具体分类,并对它进行系统阐述。在本文的实证部分中,首先对数据进行预处理,并建立多分辨小波神经网络模型,根据样本数据的特性对网络各层节点数、训练参数等进行设置,以2014年至2015年中339个交易日的上证指数为研究对象,用前311个数据对网络进行训练,用后28个数据做为测试样本,建立误差率为主的模型评价标准,对2015年股灾的波动状况进行分析和预测。结果表明,我国的上证指数并非杂乱无章,而是可预测的,存在一定的运行规律；多分辨小波神经网络对于股价测试样本的误差率小,效果优良,具有较高的推广价值。
[Abstract]:Stock market is an important field of national macro-control and direct financing of enterprises. Its income and risk are proportional to each other. Because of the characteristics of stock price random walk, and subject to economic and political changes at home and abroad. Many factors, such as the psychological characteristics of shareholders, are difficult to evaluate scientifically. The influence factors of stock price are complex and highly nonlinear, and the traditional statistical model has high noise. It is difficult to achieve ideal results for forecasting stock price with many characteristics such as nonlinearity. Therefore, the establishment of a high accuracy. The simplified and practical evaluation model is of great practical significance for the investment of the securities market and the national macro-control. At present, the majority of investors use the pattern and shape according to the stock trend for stock price prediction. Because of the large number of analysis methods, lack of theoretical support of scientific system, and the independence of each index, the prediction accuracy is not high. Artificial intelligence is a large science. Covering computer, psychology, image processing and other knowledge, in recent years has made a breakthrough and wide application, neural network is a branch of artificial intelligence, wavelet neural network is based on the traditional network. Wavelet transform is introduced to transform it, which has the characteristics of nonlinear approximation, self-organizing learning, simple structure and so on, and it also has the black box identification ability of wavelet analysis. In 2015, the stock market experienced a leveraged mad cow, a stock price limit, a limit, a suspension, and the government rushed to rescue the market. This paper first introduces the detailed process of the stock market crash in 2015, analyzes the causes of the stock crash from the macroeconomic point of view, and then introduces the relevant knowledge background. Including the basic knowledge of the stock market, the current stock price forecasting methods, neural networks and wavelet analysis related concepts, wavelet neural networks and the basic characteristics of the classification. In the empirical part of this paper, we first preprocess the data, and establish a multi-resolution wavelet neural network model, according to the characteristics of the sample data, the number of nodes in each layer of the network. The training parameters were set up, and the Shanghai Stock Exchange Index (SSE) of 339 trading days from 2014 to middle of 2015 was taken as the research object, and the network was trained with the first 311 data. The last 28 data are used as test samples to establish a model evaluation standard based on error rate to analyze and forecast the fluctuation of stock market in 2015. The results show that the Shanghai Stock Exchange Index in China is not disorderly. It is predictable, and there are certain rules of operation. Multiresolution wavelet neural network has low error rate and good effect for stock price test samples.
【学位授予单位】：华中师范大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：F832.51;TP183

【参考文献】