【作者】 刘倩；

【导师】 梁久祯；

【作者基本信息】 江南大学 ， 计算机软件与理论， 2009， 硕士

【摘要】 本文专门研究了沪深股市中的30只股票的特征和规律,从多个方面切入研究股票的性质,以期对股票序列的趋势能够有更多的了解,使广大投资者对某只股票进行投资时,有更多的依据以及对于风险的估计。文中对股票时间序列从以下三个方面进行了研究和分析:1)通过正态性检验,发现各样本股票数据均表现出不同程度的尖峰和胖尾以及左偏现象,证明了所有的样本序列全都不属于正态分布,也就是说明它们不能够适用线性方法进行分析,必须采用重标极差分析方法(R/S分析方法)的非线性方法。2)应用R/S分析方法验证他们具有长期记忆的特性,并且结合支持向量回归机(简称SVR)提高了测定平均循环周期时的精确性。在使用R/S分析方法分析股票的同时,我们主要引入了SVR方法,它是属于数据挖掘中的一种新方法,相对于传统的回归方法,它可以解决非线性回归问题的一种方法。利用SVR方法与R/S双对数曲线和V统计量曲线的结合,对平均循环周期进行测定,由于它不仅保持了原曲线的本身趋势,而且相较于原曲线来说更加的平滑,因此,大大增加了我们得到的股票时间序列平均循环周期的精确度。3)我们使用粗糙集进行股票时间序列的规律发现,规律发现的过程包括数据预处理和F-规律发现两个部分。其中数据预处理首先要对股票时间序列进行数据清洗,将清洗后的时间序列按照变化趋势的不同进行分割,完成分割后,其中的每个时间段内的变化趋势是保持不变的,于是时间序列就转换成为了一系列静态模式,里面的每一种模式代表了一种行为趋势,而且还可以去掉它的时间依赖性。然后再把与各模式相关的属性抽取出来组成一个可以被粗集理论处理的信息表,然后采用粗集理论对信息表进行F-规律的发现。分析结果:抽取的这30只沪深股市的股票普遍具有分形特征,其变化是具有循环性的。当发现的股票风险规律大于其原始规律时,相应的股票Hurst指数值一般都小于0.59。说明了Hurst指数在0.59以下时,此种股票的稳定性比较差,当有风险属性进行入侵时,这只股票在接下来的趋势中会出现逆向波动的可能性较大。更多还原

【Abstract】 This paper gives specialized research to the features and regular pattern of 30 different stocks from Shanghai and Shenzhen stock market, and studys their inner characteristics in many aspects to acquire more and further recognition on the trend of stock share series, which enables the investors to gain more theory basis when they buy some stock and estimate its risk. This paper is expanded into three main parts to study and analyze the stock time series.1) Through the normality tests, All the sample stock-data curves show different phenomenon of Aiguille and Fat-tail, as well as Left-Deviation, which proves that none of the sample series belong to normal distribution, in other words, the linear method can’t be applied into the analysis process, so we adopted the R/S method instead.2) The application of R/S method has verified the long-term memory characteristics of the sample series, and Support Vector Regression (SVR) is combined to enhance the precision of Average circle period.Meanwhile, we mainly introduce the SVR method. It appears as a new data-mining method to take place in the traditional regression method, which can also well solve the non-linear regression problem. Through the combination of SVR method, R/S log-log type curve and V-statistics curve, Average circle period is tested, which not only maintains its own original trend of the curves, but also becomes smoother. The integrated method has greatly boosted the precision of the average circle period in the stock time series.3) We expect to find the rules of the stock time series by using rough sets, this process includes two parts, data Pretreatment and F-rule discovery.Data Pretreatment part needs to clean the stock time series data first, and then divide them into different time intervals according to their respective transformation tendency. Ensure that the trend in every time interval keeps single and changeless, thus all the time series are transformed to a series of totally static pattern, and every pattern represents a behavior trend and their time dependence can be ignored. The respective attribute of every pattern is extracted to build an information table which can be processed by the rough sets theory. Finally, we operate the information table through rough sets theory to find the F-rule.Result Analysis: The selected 30 stocks from Shanghai and Shenzhen stock markets generally have the fractal feature, their transformation courses are circulatory. When the discovered rule of stock risk is larger than its original rule, its related stock Hurst index is usually lower than 0.59, indicates that once a stock’s Hurst index is below 0.59, its stability is relatively weak, especially when encountering the invading of risk attributes, the possibility of appearing reverse fluctuation for the stock will be bigger in the following trend.更多还原