基于同辈群体分析方法的股票走势预测研究

发布时间：2018-02-25 16:37

本文关键词： 流特征模式同辈群体传动模型股票走势预测　出处：《合肥工业大学》2014年硕士论文　论文类型：学位论文

【摘要】：股票走势的预测对投资决策具有重要的指导作用。然而,股票走势受到宏观经济政策、突发事件以及股市人为操控等因素的影响,使股票走势难以有效预测。论文针对股票走势的突变性和多变性问题,提出了基于流特征模式的股市跟踪预测算法(SFM-PG)和基于传动模型的股市预测算法(TM-PG),分别解决股票.短期走势和中期走势的预测问题,具体研究工作如下：(1)分析了将马尔科夫毯模型和同辈群体分析方法相结合的可行性。传统的同辈群体算法使用的是平均权重法对目标对象的同辈群体建立跟踪模型,论文使用加权平均法建立目标对象的跟踪模型,能够使同辈群体中的成员更加紧密的跟踪自己的目标。(2)提出一种基于马尔可夫毯和同辈群体的股市预测算法(SFM-PG).SFM-PG算法选取目标股票的马尔科夫毯作为其同辈群体,根据同辈群体之间的接近度,给出一种窗口跟踪式预测模型,通过对同辈群体权重的动态更新进行跟踪式预测,减少股票数据分布非正态性对预测的影响；进而,使用滑动窗口提取时序数据中的特征并形成流特征,通过与模式知识库的匹配提取流特征模式,并利用与流特征模式对应的知识调整预测结果,以减少由于突变所引入的预测误差。针以上证行业板块股票的实验结果表明该算法具有较高的预测精度。(3)在SFM-PG算法实证分析获得实用性和有效性的基础上,结合同辈群体算法和行为传动性的思想,提出了TM-PG算法,对股票中期预测问题和综合指数走势行为之间的传动性进行了研究,对于股票的预测提供了另一种思路。对股票综合指数进行了实证分析,结果表明在中期预测上,该算法具有较好的预测精度,同时验证了综指之间在走势变化时具有传动性。
[Abstract]:The prediction of stock trends plays an important guiding role in investment decision-making. However, stock trends are influenced by macroeconomic policies, emergencies and artificial manipulation of the stock market. It is difficult to predict the stock trend effectively. A stock market tracking forecasting algorithm based on flow characteristic mode (SFM-PGG) and a transmission model-based stock market prediction algorithm (TM-PGN) are proposed to solve the forecasting problems of stock, short term trend and medium term trend, respectively. The specific research work is as follows: (1) the feasibility of combining Markov blanket model with peer group analysis method is analyzed. The traditional peer group algorithm uses the average weight method to establish a tracking model for the peer group of the target object. In this paper, the method of weighted average is used to establish the target object tracking model. This paper proposes a stock market prediction algorithm based on Markov blanket and peer group. SFM-PGN. SFM-PG algorithm selects the Markov blanket of the target stock as its peer group. According to the closeness between peer groups, a window tracking prediction model is presented, which can reduce the influence of non-normal distribution of stock data on prediction by tracking the dynamic update of peer group weight. The sliding window is used to extract the features from the time series data and form the flow features. The flow feature patterns are extracted by matching the pattern knowledge base, and the prediction results are adjusted by using the knowledge corresponding to the flow feature patterns. In order to reduce the prediction error caused by sudden change, the experimental results of the stock market in Shanghai Stock Exchange show that the algorithm has a high prediction accuracy. Based on the empirical analysis of the SFM-PG algorithm, it has been proved to be practical and effective. Combined with the idea of peer group algorithm and behavior transmission, the TM-PG algorithm is proposed to study the transmission between the stock medium-term forecasting problem and the behavior of composite index trend. An empirical analysis of the stock composite index shows that the algorithm has better prediction accuracy in the medium term forecasting, and it also verifies that the composite index is drivable when the trend of the composite index changes.
【学位授予单位】：合肥工业大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：F832.51;F224

【相似文献】