基于XGBoost算法的上证指数预测方案设计研究

发布时间：2018-04-06 02:20

本文选题：数据挖掘　切入点：上证指数　出处：《上海师范大学》2017年硕士论文

【摘要】：数据挖掘技术产生于20世纪80年代后期,90年代有了突飞猛进的发展,随着技术的不断成熟,越来越多的学者将其广泛运用于不同的领域;其中,与金融领域的结合能够给广大投资者带来额外收益;股票市场是一个受多方信息影响的复杂系统,股市的涨跌由于其高度不稳定性,更是难以预测。投资者面对大量的股市信息,通常希望能够利用已知的历史信息运用某种方式对未来的市场涨跌进行预测,以应用于投资,获得超额收益。面对巨大的信息量,人工进行处理显然不现实:花费的成本也过于昂贵;所以有许多学者运用例如支持向量机、BP神经网络等机器学习的方法来对股市的涨跌进行预测;这一领域逐渐成为近两年待解决的热点问题;但是支持向量机等方法有一定的局限性,为了达到最优的分类效果,要采用高纬度的平面进行分类,这无疑增加了模型的复杂度;XGBoost算法作为2015年新提出的算法,具有运算效率和准确率高的优点,所以作者运用这一新的算法对股市涨跌进行预测,为投资者提供一种新的投资决策有效性方案。本文结合国内股票市场和国际上主要的股票指数,运用了支持向量机、决策树模型和XGBoost算法对上证综指、上证50指数、标准普尔指数的涨跌进行预测;同时,为了尽可能提高支持向量机、决策树和XGBoost算法对股市涨跌预测的效果,作者还把与成交量有关的数据进行了处理,使它的数值与其他指标相差不是太大;与此同时,还将XGBoost算法的有关参数进行了调优。选取了 28个技术指标作为输入变量,将预测的第二天的股市涨跌作为分类的输出变量;利用RStudio软件进行支持向量机、决策树和XGBoost建模,并得到了相对合理的实证结果,结果显示XGBoost模型对上证综指有非常理想的预测效果,预测的准确率达到了 70%以上,这与XGBoost算法的原理有关,它迭代每次的误差,达到最小化平方损失函数,所以比普通算法的准确率要高;上证50和标准普尔指数的预测准确率达到了 60%到65%,这可能与这两个指数只是选取的一部分股票作为样本有关;按照趋势进行划分时,也能够得到更高的预测准确率,运用XGBoost算法的预测结果进行投资,结果也显示能够使投资者获得理想的超额收益,支持向量机和决策树略低,也达到了 60%以上。可以看出,机器学习方法对股市预测和投资有一定的指导意义。给投资者的决策和政府监管提供了一个方便,切实可行的方案。
[Abstract]:Data mining technology emerged in the late 1980s and 1990s with the rapid development of technology, with the continuous maturity of the technology, more and more scholars widely used it in different fields.The combination with the financial field can bring extra income to the majority of investors. The stock market is a complex system affected by many kinds of information, the stock market's rise and fall is more difficult to predict because of its high instability.Faced with a large amount of stock market information, investors usually hope to use known historical information to predict the future market fluctuations in a certain way, in order to apply it to investment and obtain excess returns.Faced with the huge amount of information, it is obviously not realistic to deal with it manually: the cost is too high, so many scholars use machine learning methods such as support vector machine (SVM) and BP neural network to predict the stock market's rise and fall.This field has gradually become a hot issue to be solved in the last two years. However, support vector machine and other methods have some limitations. In order to achieve the optimal classification effect, the high-latitude plane should be used for classification.This undoubtedly increases the complexity of the model and the XGBoost algorithm, as a new algorithm proposed in 2015, has the advantages of high computational efficiency and high accuracy, so the author uses this new algorithm to predict the stock market's ups and downs.To provide investors with a new investment decision effectiveness scheme.This paper uses support vector machine, decision tree model and XGBoost algorithm to forecast the rise and fall of Shanghai Composite Index, Shanghai Stock Exchange 50 Index and Standard & Poor's Index.In order to improve the prediction effect of support vector machine, decision tree and XGBoost algorithm on stock market fluctuation, the author also processed the data related to trading volume so that its value is not too different from other indexes; at the same time,The parameters of XGBoost algorithm are also optimized.28 technical indexes are selected as input variables, and the stock market fluctuation in the second day of forecast is taken as the output variable of classification. The support vector machine, decision tree and XGBoost are used to model the model using RStudio software, and a relatively reasonable empirical result is obtained.The results show that the XGBoost model has a very good prediction effect on the Shanghai Composite Index, and the prediction accuracy is over 70%, which is related to the principle of the XGBoost algorithm. It iterates the error every time to minimize the square loss function.So it's more accurate than the normal algorithm; the accuracy of the Shanghai 50 and Standard & Poor's indices is between 60% and 65%, which may be related to the fact that the two indices are only selected as a sample; when they are divided according to the trend,It can also get higher prediction accuracy, using the prediction results of XGBoost algorithm to invest, the results also show that the investors can get ideal excess returns, support vector machines and decision trees are slightly lower, up to more than 60%.It can be seen that the machine learning method has certain guiding significance for stock market prediction and investment.It provides a convenient and feasible scheme for investors'decision making and government regulation.
【学位授予单位】：上海师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F831.51

【参考文献】