PCA和随机森林在BARRA量化对冲模型中的应用研究

发布时间：2018-02-12 17:00

本文关键词： 随机森林 PCA BARRA模型量化对冲　出处：《西安科技大学》2017年硕士论文　论文类型：学位论文

【摘要】：近几年,随着国内上市公司越来越多,同时伴随着市场的大起大落,使人们意识到传统定性投资方法的局限性。因此,国外业绩较为突出的量化投资策略引起了国内投资者的关注。本文基于BARRA的理论框架,进行了如下研究:(1)首先,基于BARRA结构化多因子风险模型,构建了本文的纯因子检验模型和BARRA量化对冲模型。(2)针对BARRA量化对冲模型中多个同类因子等权合成从而导致重要信息不突出的问题,给出了使用随机森林获得多个因子的重要程度,进而将其作为因子权重的方法。首先,结合最大化稳健收益的策略目标,给出了样本的分类属性;进而结合对冲模型需要选择风格暴露方向的特点,给出了分样本集学习以确定因子权重的方法。实证结果表明,使用该方法所得的因子权重相比等权处理方法,,具有明显的改进效果。(3)针对BARRA量化对冲模型中加权合成的风格因子之间存在信息重叠进而导致因子收益率估计不准确的问题,给出了使用PCA的方法在同类因子中进行信息提取,以增强模型参数估计准确度的思路。具体是将一类风格中的若干因子进行PCA处理,综合考虑各成分的信息贡献率以及与其它风格因子之间的线性相关性,从而选择一个主成分作为该类风格因子的替代。实证结果表明,这一处理方法,虽然会使某些风格损失部分有价值的信息,但模型整体估计的准确度有所提升,从而为对冲模型风格暴露的设置提供了更可靠的依据。(4)鉴于多因子模型的实现和测试环节众多、容易出错的问题,本文设计并实现了一套完整的多因子量化选股系统。通过试运行,结果表明该系统提高了模型测试与应用的准确度和效率,为投资者提供了有效的选股工具。
[Abstract]:In recent years, with more and more domestic listed companies, and accompanied by the ups and downs of the market, people are aware of the limitations of traditional qualitative investment methods. The quantitative investment strategy with outstanding performance abroad has attracted the attention of domestic investors. Based on the theoretical framework of BARRA, this paper makes the following research as follows: 1) first, based on the structured multi-factor risk model of BARRA, The pure factor test model and the BARRA quantitative hedging model are constructed in this paper. Aiming at the problem of equal-weight synthesis of multiple similar factors in the BARRA quantitative hedging model, the important information is not prominent. In this paper, the importance of using random forest to obtain multiple factors is given, and then it is regarded as the weight of factors. Firstly, the classification attributes of samples are given in combination with the strategic goal of maximizing robust returns. Furthermore, considering the characteristic that hedging model needs to choose the direction of style exposure, the method of learning by sample set to determine the weight of factor is given. The empirical results show that, Compared with the equal weight processing method, the factor weight obtained by this method has obvious improvement effect. (3) aiming at the problem of information overlap among the weighted composite style factors in the BARRA quantitative hedging model, which leads to inaccurate estimation of the factor return rate. The idea of using PCA to extract information from similar factors to enhance the accuracy of model parameter estimation is given. In particular, some factors in a class of styles are processed by PCA. Considering the information contribution rate of each component and the linear correlation between each component and other style factors, a principal component is selected as the substitution of this kind of style factor. Although some styles may lose some valuable information, the accuracy of the overall estimation of the model has improved, thus providing a more reliable basis for hedging model style exposure settings.) given the numerous steps in the implementation and testing of multi-factor models, In this paper, a complete multi-factor quantitative stock selection system is designed and implemented. The results show that the system improves the accuracy and efficiency of model testing and application, and provides an effective stock selection tool for investors.
【学位授予单位】：西安科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F832.51;O212.4;TP311.52

【参考文献】