基于数据挖掘的金融时间序列预测研究与应用

发布时间：2018-03-09 02:01

本文选题：数据挖掘　切入点：金融时间序列　出处：《江南大学》2015年硕士论文　论文类型：学位论文

【摘要】：获得1997年诺贝尔经济学奖的罗伯特·默顿(Robert Merton)认为,资金的时间价值、资产定价与风险管理构成了现代金融理论的基石,如何在不确定的环境下对资源进行跨期的最优配置是其最重要的问题。随着计算机软件的发展,各种软件和应用得到广泛的运用,并持续不断的融入各种商业、社会生产和生活等。金融是社会经济中极其重要的组成部分。对金融数据进行数据挖掘,就是在大量非稳态、非线性和高信噪比的不确定的金融数据中挖掘出有价值的信息。数据挖掘技术在金融领域的研究日渐兴起。虽然传统的数据挖掘技术在处理常见数据时表现尚可,但是处理非稳态、非线性和高信噪比的不稳定的金融时间序列时表现出一定的局限性。因此改进现有数据挖掘技术方法在金融时间序列中的应用研究显得尤为重要。针对这一问题,本文以数据挖掘中的聚类算法为切入点,挖掘出金融时间序列中潜在的知识并形成模型进行预测,对金融时间序列展开了深入的研究,主要工作如下:首先,针对DBSCAN聚类算法无法处理变化密度数据集的问题,结合初始点优化和参数自适应的方法改进DBSCAN算法,提出了一种新的可应对变化密度数据集的基于密度的空间聚类算法(OS-DBSCAN)。实验结果表明新的改进的基于密度的空间聚类算法能够针对变化密度数据集进行聚类,并能在给予初始参数之后根据数据集自身的特征和属性进行参数自适应,而且相对于传统的DBSCAN算法,初始点优化与参数自适应的密度空间聚类算法能够在一定程度上提高聚类的质量。其次,针对基于数据挖掘的金融时间序列预测的问题,结合所提出的OS-DBSCAN聚类算法和基于粒子群优化的SVR回归预测算法,提出了一种可应对非稳态、非线性和高信噪比的金融时间序列的混合算法。使用真实金融数据,如Microsoft与Intel股票数据、TWII金融指数数据和A股2259只个股历史交易数据,对算法展开了深入研究。实验数据显示,基于OS-DBSCAN和粒子群优化的SVR的混合算法在对金融时间序列预测中表现出一定的优势。相对于传统的处理方法,基于OS-DBSCAN和粒子群优化的SVR算法在股票价格和金融指数的回归预测实验中获得了较好的准确率,能够提高对股票次日涨幅预测的准确率。数据挖掘算法在金融时间序列预测上取得了一定的效果和实用性。
[Abstract]:Robert Merton, who won the Nobel Prize in economics in 1997, argues that the time value of capital, asset pricing and risk management form the cornerstone of modern financial theory. With the development of computer software, various kinds of software and applications have been widely used and continuously integrated into various businesses. Social production and daily life. Finance is an extremely important part of social economy. The data mining of financial data is in a large amount of unsteady state. Valuable information is mined from uncertain financial data that is nonlinear and high SNR. Data mining technology is emerging in the financial field. Although traditional data mining techniques perform well in processing common data, But dealing with unsteady states, Nonlinear and unstable financial time series with high signal-to-noise ratio (SNR) show some limitations. Therefore, it is very important to improve the application of existing data mining techniques in financial time series. Based on the clustering algorithm in data mining, this paper excavates the potential knowledge in financial time series and forms a model to predict it. The main work of this paper is as follows: first, Aiming at the problem that the DBSCAN clustering algorithm can not deal with the variable density data set, the DBSCAN algorithm is improved by combining the initial point optimization and parameter adaptive method. A new density based spatial clustering algorithm is proposed, which can deal with the variable density data sets. The experimental results show that the new density based spatial clustering algorithm can be used to cluster the variable density data sets. After giving the initial parameters, the parameters can be adaptively adapted according to the characteristics and attributes of the data set, and compared with the traditional DBSCAN algorithm, The initial point optimization and parameter adaptive density space clustering algorithm can improve the clustering quality to some extent. Secondly, aiming at the problem of financial time series prediction based on data mining, Combined with the proposed OS-DBSCAN clustering algorithm and the SVR regression prediction algorithm based on particle swarm optimization, a hybrid algorithm for unsteady, nonlinear and high signal-to-noise ratio (SNR) financial time series is proposed. For example, the Microsoft and Intel stock data and the TWII financial index data and the historical trading data of 2259 A-share stocks are studied in depth. The experimental data show that, The hybrid algorithm based on OS-DBSCAN and particle swarm optimization (PSO) SVR has some advantages in financial time series prediction. The SVR algorithm based on OS-DBSCAN and particle swarm optimization is applied to the regression prediction experiment of stock price and financial index. It can improve the accuracy of forecasting the next day's increase of stock. The data mining algorithm has achieved certain effect and practicability in forecasting financial time series.
【学位授予单位】：江南大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：F830;TP311.13

【参考文献】