中国主要股指的分形分析与BP神经网络预测

发布时间：2018-01-22 11:10

本文关键词： 沪深300指数 R/S分析 BP神经网络预测　出处：《大连理工大学》2013年硕士论文　论文类型：学位论文

【摘要】：众所周知,股票市场在国民经济中的地位是极其重要的。中国股市到底遵循什么样的规律呢?是线性的还是非线性的,是随机游走模型抑或非高斯分布的无记忆模型还是一个分形结构呢?只有把这些问题彻底搞明白了,才能更好的分析和预测市场、才能抓住市场的规律,才能更好的利用我国的资本市场。中国的股票市场包含了2000多家上市公司,以及各种板块、指数等,对每一个都进行分析是很难实现的。在实际的分析与研究中,一般会选择上证指数和深证成指作为代表进行分析,因为上证指数反映了上海证券交易所的整体走势、深证成指反映了深圳证券交易所的整体走势。本文将沪深300指数和上证指数、深证成指分别叠加分析,发现沪深300指数和上证指数、深证成指的走势基本一致,可以说沪深300指数的走势反映了中国股票市场的整体走势,因此选用了沪深300指数来分析中国的股票市场。在分析中国的股市是不是随机游走时采用的是重标极差的分析方法,根据沪深300指数的Hurst指数和0.5的大小比较来进行确定。通过比较发现沪深300指数的日线、周线和月线的Hurst指数均大于0.6,从而说明中国的股市不是随机游走的,而是有记忆性的非线性结构,并且是一个分形结构。从而也说明中国股市的有效性不强。并根据重标极差分析可确定论文第四章较为合理的预测区间。在预测沪深300指数的收盘价时,本文是结合技术分析,应用BP神经网络进行预测,作为基本模型。在基本模型的基础上,将预测结果和EMA均线进行线性加权得到改进模型。这两个模型都可以跑赢大盘,改进模型精度更高。对411天数据的预测中,最终实现正数中最大偏离误差为4.69%,负数中最大偏离误差为-3.73%,预测值偏离实际值的偏差的绝对值的平均值为1.13%。将前20天的预测结果和后20天的预测结果进行比较,发现对时间序列的学习样本进行及时的数据更新有利于提高预测精度。本文最终得出三大结论：一,基于BP神经网络对沪深300指数的预测在一定程度上是可行的,可以跑赢大盘；二,选择合适的预测区间是很重要的；三,中国的股票市场是一个分形结构,其有效性不强。事件驱动型明显,将新闻、消息具体量化并加入到影响因素中是很重要的,通过具体数据成功预测股市还有很长的道路要走。
[Abstract]:As we all know, the position of the stock market in the national economy is extremely important. What kind of law does the Chinese stock market follow? Is it linear or nonlinear? is it a random walk model or a memoryless model of non-#china_person0# distribution or a fractal structure? Only by thoroughly understanding these problems can we better analyze and predict the market, grasp the laws of the market, and make better use of the capital market of our country. China's stock market contains more than 2000 listed companies, as well as a variety of plates, indices, and so on, each of which is difficult to achieve. In the actual analysis and research. Generally will choose the Shanghai Stock Exchange Index and Shenzhen Stock Exchange as the representative of the analysis, because the Shanghai index reflects the overall trend of the Shanghai Stock Exchange. The Shenzhen Composite Index reflects the overall trend of the Shenzhen Stock Exchange. This paper analyzes the CSI 300 index and the Shanghai Stock Exchange index respectively and finds the CSI 300 Index and the Shanghai Stock Exchange Index respectively. The Shenzhen Composite Index is basically in line with the trend. It can be said that the trend of the CSI 300 index reflects the overall trend of the Chinese stock market, so the CSI 300 Index has been chosen to analyze the Chinese stock market. In the analysis of whether the Chinese stock market is a random walk, the method of rescaling extreme difference is used. According to the Hurst index of Shanghai and Shenzhen 300 index and the size of 0. 5 to determine, through the comparison found that the daily line of the Shanghai and Shenzhen 300 index, the Hurst index of the week line and monthly line are all greater than 0. 6. It shows that China's stock market is not random walk, but a memory of the nonlinear structure. And it is a fractal structure, which also shows that the validity of Chinese stock market is not strong. According to the rescaled range analysis, we can determine the more reasonable prediction interval in Chapter 4th of this paper. In predicting the closing price of CSI 300 index, this paper uses BP neural network to predict the closing price of CSI 300 index, which is based on the basic model. The prediction results and the EMA mean line are weighted linearly to get the improved model. Both models can outperform the market, and the improved model has higher precision. The maximum deviation error of positive number and negative number is 4.69% and -3.73% respectively. The average absolute value of the deviation from the actual value is 1.13.The results of the first 20 days are compared with those of the next 20 days. It is found that timely updating of time series learning samples is helpful to improve prediction accuracy. This paper finally draws three conclusions: first, the prediction of Shanghai and Shenzhen 300 index based on BP neural network is feasible to some extent and can outperform the market; Secondly, it is very important to choose the appropriate prediction interval. Third, China's stock market is a fractal structure, its effectiveness is not strong, event-driven obviously, it is very important to quantify and add news and information to the influence factors. There is still a long way to go to successfully predict the stock market through specific data.
【学位授予单位】：大连理工大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：F832.51;TP183

【参考文献】