基于SVM_AdaBoost模型的股票涨跌实证研究

发布时间：2018-02-15 14:25

本文关键词： 股价预测 SVM模型 AdaBoost算法 SVM_AdaBoost强分类器　出处：《华南理工大学》2013年硕士论文　论文类型：学位论文

【摘要】：股票市场不仅是国家经济的晴雨表，在中国更是企业融资和广大股民投资的重要手段。股票涨跌能够得到大致预测，就能为投资者提供有效的股票投资指导，从而有效克服盲目投资的弊端和有效规避了股票的投资风险。本文将股票的未来走势分别划分为两种（涨、跌），把股市的波动预测转化为两类分类问题，通过股票上周数据指标来对下周股价涨跌进行预测，分别利用核函数为高斯核的最小二乘支持向量机（LSSVM）模型以及LSSVM和AdaBoost算法混合模型（SVM_AdaBoost）进行分类识别。本文首先介绍了股市的相关背景知识，国内外对于股票预测的研究综述，对几种现有的股票预测方法进行了介绍。接着全面介绍了统计学习理论和建立在其基础上的SVM方法，阐述了LSSVM的原理及其优点。并选取万科A股作为实证对象，验证LSSVM在整体预测，以及上涨市、下跌市、震荡市的预测情况，实证表明股价在单边上涨或者下跌的时候，LSSVM具有较好的分类性能。为提高震荡市以及整体市场的预测性能，引入AdaBoost算法。介绍了AdaBoost算法的理论和算法流程，指明该算法可以和SVM模型组合的优势：SVM_AdaBoost模型可以减少SVM选择核参数的复杂度，而核参数的多样性也使得SVM提供多样性的弱分类器成为可能。将支持向量机（SVM）模型作为AdaBoost算法的弱分类器，得到SVM_AdaBoost强分类器，作为识别股票周涨跌的模型。针对多样性的基分类器的构造，，本文基于模糊均值聚类算法，给出了一种基分类器的评价方法，通过这种评价方法自适应调整SVM中的核参数，从而得到一组适当精度并具有多样性的SVM弱分类器。实证结果表明，SVM_AdaBoost模型提升SVM的预测精度并优化了算法学习效率，还能提高不平衡数据中少数类的预测精度。相信随着今后进一步研究，应用SVM_AdaBoost模型进行股市预测将得到更好的效果。
[Abstract]:The stock market is not only a barometer of the national economy, but also an important means of enterprise financing and investment by the majority of investors in China. If stocks can be roughly predicted, they can provide effective investment guidance for investors. In order to overcome the disadvantage of blind investment and avoid the investment risk of stock effectively, this paper divides the future trend of stock into two kinds: rising, falling, converting the fluctuation forecast of stock market into two kinds of classification problems. The index of stock last week is used to predict the stock price's rise and fall next week. The LSSVM model with Gao Si kernel and the mixed model of LSSVM and AdaBoost algorithm are used to classify and identify the stock price. This paper first introduces the relevant background knowledge of stock market, summarizes the research on stock forecast at home and abroad, introduces several existing stock forecasting methods, and then introduces the statistical learning theory and SVM method based on it. This paper expounds the principle and advantages of LSSVM, and selects Vanke A shares as the empirical object to verify the overall forecast of LSSVM, as well as the forecast of rising market, falling market and concussion market. The empirical results show that the stock price has better classification performance when the stock price rises or falls unilaterally. In order to improve the prediction performance of shock market and the whole market, AdaBoost algorithm is introduced. The theory and algorithm flow of AdaBoost algorithm are introduced. It is pointed out that this algorithm can be combined with the SVM model to reduce the complexity of SVM's selection of kernel parameters by using the: SVM AdaBoost model. The diversity of kernel parameters makes it possible for SVM to provide multiple weak classifiers. The support vector machine (SVM) model is used as the weak classifier of AdaBoost algorithm to obtain SVM_AdaBoost strong classifier. As a model to identify the rise and fall of stock cycle, this paper presents an evaluation method of base classifier based on fuzzy mean clustering algorithm, which adaptively adjusts the kernel parameters in SVM. A set of SVM weak classifiers with proper accuracy and diversity are obtained. The empirical results show that the SVMAdaBoost model can improve the prediction accuracy of SVM and optimize the learning efficiency of the algorithm, as well as the prediction accuracy of a few classes in unbalanced data. It is believed that the application of SVM_AdaBoost model in stock market forecasting will be more effective with further research in the future.
【学位授予单位】：华南理工大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：F832.51;F224;TP18

【参考文献】