【作者】 隋学深；

【导师】 齐中英；

【作者基本信息】 哈尔滨工业大学 ， 技术经济及管理， 2008， 博士

【摘要】 对股票市场价格行为进行研究,在宏观和微观方面都有重要的现实意义。从宏观角度来看,深刻理解股票市场的价格行为的波动特征,是政府制定干预市场政策的基础;从微观角度来看,则影响到市场参与者(包括投资者)的市场投资策略与手段。基于模型分析是研究股票市场价格行为的传统方法,该法试图通过对时序数据进行分析,完成对时序系统的预测、建模和控制。其主要采用演绎研究方法,推理遵循严格的逻辑,但问题在于其前提假设无法得到验证,而且该方法试图建立一个全局的普适性的模型。股票市场作为复杂系统的行为模式使得对其建立全局、精确的数学模型的可能性大大降低,并且对系统进行完全意义上的预测已不可能。将数据挖掘的思想引入到股票市场价格行为分析中,对股票市场时序数据进行挖掘,从中发现蕴含的系统规律,将其用于股票市场价格行为的分析和预测,这将很好地弥补原有模型分析方法的不足,为股票市场价格行为问题的研究提供了一种新的思路和手段。本文将基于数据挖掘方法对股票市场价格行为进行如下几个方面的研究:首先,股票市场多尺度技术指标提取研究。本研究根据股票市场时间序列具有信噪比低、多尺度性、非线性、非平稳性和长记忆性等特征,构建了一套股票市场多尺度技术指标提取方法。该方法在去噪的基础上,首先根据股票市场数据的多尺度特征,应用小波多尺度分析得到多尺度下的股票市场时序数据;然后,对多尺度时序数据进行相关性分析以提取其长记忆性特征;最后,应用单位根检验和滑动平均方法来进行趋势分析,以提取其非平稳特征。其次,股票市场技术指标体系构建研究。根据股票市场时间序列分类边界复杂性高的特点,本文选用了基于分类复杂度的属性约简方法做为股票市场技术指标体系的构建方法。为了便于对股票市场趋势预测效果进行比较研究,在构建股票市场多尺度技术指标体系的同时,本文也构建了股票市场传统技术指标体系。再次,股票市场趋势预测及其预测机理研究。在对股票市场进行可预测性分析的基础上,分别对本文构建的多尺度技术指标体系和股票市场传统技术指标体系进行趋势预测研究。预测的实证研究表明,和传统方法相比,本文构建的多尺度技术指标体系和预测方法使得股票市场预测准确率提高了十几个百分点。最后,为了解释本研究预测准确率提高的机理,应用模糊粗糙集方法对两种体系技术指标对股票市场价格行为趋势预测能力进行了分析,揭示出了股票市场尺度技术指标具有优秀预测能力的客观机理。最后,基于多尺度技术指标的股票市场价格行为技术交易规则挖掘研究。介绍了股票市场技术分析中常规的技术交易规则,并在此基础上提出了基于数据挖掘的股票市场价格行为技术交易规则概念。本文分别选择了基于粗糙集理论和决策树的技术交易规则挖掘方法对第三章构建的多尺度技术指标体系进行了股票市场价格行为的多尺度技术交易规则挖掘。以决策树方法挖掘的多尺度技术交易规则为例,结合股票市场技术分析的投资实践经验,揭示了多尺度技术指标组合形成的技术交易规则对股票市场趋势价格行为的预测机理。更多还原

【Abstract】 The research on stock market price behaviors is important in macro-scope and microcosmic aspect. In the view of macro-scope, realizing the stock market behavior character deeply is a foundation of government to establish stock market police. In the view of microcosmic, realizing the stock market behavior character deeply has impact on the investment strategy of the stock market participator. Model analysis is a traditional method on stock market research. By analysis the time series data, it could accomplish the forecasting, establishing model and controlling of the time series system. Following the strict logic, it use induction research method. The problem is it can not validate the precondition and it try to establish a universal model. There is little probability to establish a universal and precise model for stock market as a complexity system and it impossible to make whole forecasting. Use data mining method to analysis stock market behavior can fetch up the model method shortage. This work will research following problem based on data mining method.First, research on stock market multi-scale technical index induced. According to low I/N rate, multi-scale, nonlinear, non-stationary and long memory and so on characters of stock market, we establish a stock market multi-scale technical index induced method. We use wavelet transform to get multi-scale time series data of stock market. For the multi-scale time series, we use correlation analysis to get the long memory character. We use unit root and moving average methods to get the non-stationary character.Second,research on establishment of stock market multi-scale technical index system. For the complexity of stock market time series neighbor, we use attribute reduction which based on classification complexity as stock market technical index system establish method. To compared with the result of multi-scale technical index system, we also establish traditional technical index system by the above four methods.Third, research on stock market price behavior trend forecasting and the forecasting mechanism. Based on the forecasting feasibility, we forecast stock market price behavior trend with multi-scale technical index system and traditional technical index system respectively. Compared with the result of traditional technical index system, the correct forecasting rate of multi-scale technical index system is almost twenty percent better. To interpret the mechanism of the better forecasting, we use fuzzy rough set method to analysis the forecasting ability of the multi-scale technical index and traditional technical index.Finally,research on stock market price behavior technical trade rule of multi-scale technical index system. Based on the introducing stock market traditional technical trade rule, we propose stock market price behavior technical trade rule on data mining. We induce multi-scale technical index trade rule with rough set and decision tree methods. As the example of decision tree methods, combined with the experience of technical analysis investment, we interpret the mechanism of stock market multi-scale technical index trade rule with fuzzy rough set.更多还原