基于混合核函数支持向量机的基金投资决策研究

发布时间：2018-03-25 07:34

本文选题：支持向量机　切入点：核函数　出处：《华南理工大学》2012年硕士论文

【摘要】：证券投资基金发展到今天，已成为世界范围内的重要的投资渠道和理财工具，截至2011年末全球共同基金的规模已经达到25.92万亿美元。近年来我国基金也迅速发展，2011年末我国基金数量突破800支，资产净值增长迅速。我国基金快速发展的同也凸显出了一些问题。2008年金融危机席卷全球后，我国股市一直处于较为低迷的状态，而基金的业绩也处于下滑状态，2011年我国末我国66家基金公司资产管理净值缩水15.16%。在当下基金业普遍不景气的情况下，如何选择具有投资价值的基金成为亟待解决的问题。根据目前基金投资面临的实际问题，结合我国对基金绩效评价体系的不足，本文提出了基于混合核函数支持向量机的基金投资决策模型。首先建立混合核函数支持向量机，是通过高斯核函数和多项式核函数的线性组合，建立新的核函数，以改善高斯核函数的泛化能力；同时对高斯核本身进行改进，建立多宽度高斯核函数，能够对输入指标进行不同的赋值，以提高模型的分类精度。其次本文提出了基于遗传算法的SVM参数优化方案，并且针对非平衡数据建模，引入F测度作为适应度函数，进一步提高模型的预测精度。最后本文综合了目前国内外金融机构的基金绩效评价指标体系，提出了涵盖收益、风险、基金经理投资管理能力和基金投资组合等方面的综合指标体系。最后本文应用2008年-2011年间的39支股票型基金进行实证分析，得出以下结论：1、通过引入F测度，可以改善非平衡数据的预测样本精度，特别是对于少数类样本的预测精度会有大幅度的提升；2、其次不同的核函数对于分类的精度有较大的影响，混合核函数支持向量机模型的预测分类精度较其他3个模型均有一定程度的提升；3、本文针对2008年后我国基金出现的大范围收益率为负的现象，给出了不同的投资选择标准，并应用本文提出的模型进行实证分析，，认为该投资标准能够帮助投资者在行情较差的情况下，选取亏损相对较少的基金；4、本文给出了基金投资决策的一般步骤。
[Abstract]:Today, securities investment funds have become an important investment channel and financial tool in the world. By the end of 2011, the size of global mutual funds had reached US $25.92 trillion. In recent years, China's funds have also developed rapidly, with the number of Chinese mutual funds exceeding 800 at the end of 2011. The rapid development of Chinese funds also highlights some problems. After the 2008 financial crisis swept the world, China's stock market has been in a relatively low state. At the end of 2011, the net asset management net value of 66 fund companies in our country shrank 15.16%. In the current situation of the general depression of fund industry, how to choose the fund with investment value becomes an urgent problem to be solved. According to the actual problems faced by fund investment and the deficiency of fund performance evaluation system in China, this paper proposes a fund investment decision model based on hybrid kernel function support vector machine. Firstly, a hybrid kernel function support vector machine (HSVM) is established. Through the linear combination of Gao Si kernel function and polynomial kernel function, a new kernel function is established to improve the generalization ability of Gao Si kernel function. It can assign different input indexes to improve the classification accuracy of the model. Secondly, this paper proposes an optimization scheme of SVM parameters based on genetic algorithm, and introduces F measure as fitness function for non-equilibrium data modeling. Finally, this paper synthesizes the current evaluation index system of fund performance of domestic and foreign financial institutions, and puts forward the coverage of income, risk, Fund manager investment management ability and fund portfolio and other aspects of the comprehensive index system. Finally, this paper applies 39 equity funds from 2008 to 2011 to carry out empirical analysis, and draws the following conclusion: 1. By introducing F measure, we can improve the accuracy of forecasting sample of unbalanced data. Especially for a small number of samples, the prediction accuracy will be greatly improved, followed by different kernel functions have a greater impact on the accuracy of the classification. The prediction accuracy of the hybrid kernel support vector machine model is improved to some extent compared with the other three models. In this paper, different investment selection criteria are given in view of the negative rate of return on a large range of funds in China after 2008. By using the model proposed in this paper, the author thinks that the investment standard can help investors select the fund with relatively little loss in the case of poor market. This paper gives the general steps of the investment decision of the fund.
【学位授予单位】：华南理工大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：F224;F832.5

【参考文献】