论文总字数:45930字
摘 要
量化投资,即用数量化的方法进行投资,运用统计、优化等数学方法建立模型,并用计算机语言进行实现。多因子模型是一种重要的量化选股模型,它是基于套利定价理论发展出来的完整的风险模型。本文尝试结合统计学方法构建符合A股市场特征的多因子选股模型。
因子选取是多因子模型中非常关键的一点,本文首先选取了38个基础因子,包括估值、盈利能力、成长性、营运质量、流动性和技术反转等七大类因子。其次,利用2007年12月至2014年12月共7年的全A股数据,运用因子IC值法和相关关系法等进行因子的有效性检验和冗余性检验,最终得到了用于构建多因子选股模型的8个有效因子。
接着本文使用打分法、主成分分析法、回归法、K-Means聚类法构建了6种多因子选股策略,并使用2015年1月至2017年12月的数据进行五分位分层回测。回测中发现,除了主成分分析法,其他5种多因子策略组合每组间区分度都较好,且top组都明显跑贏了沪深300指数。综合分析年化收益、夏普比率、超额胜率、最大回撤等业绩评价指标,K-Means聚类等权打分法在6种多因子选股策略中表现最优,年化超额收益达到23.515%,超额胜率也大于60%。
当然本文的模型也存在一些不足,如回测中回撤较大,对因子挖掘的深度和广度还不充分,这也将是我们进一步研究的方向。
关键词:多因子选股,K-Means聚类,回归分析,主成分分析
Abstract
Quantitative investment is using quantitative methods to invest, which takes financial products, such as stock, commodity futures and stock index futures as investment targets. It uses statistics and optimization model to capture the trend of asset price and thereby transforms it into profit. The multiple-factor model is an important quantitative model, which is a complete risk model based on APT. This paper attempts to construct the multiple-factor stock selection model using statistical methods.
Factor selection is the key point in multiple-factor model. Firstly, this thesis selects 38 factors which may influence the stock returns, including valuation, profitability, growth, operation quality, liquidity and contrarian. Secondly, we use IC value method and correlation method to test factors based on Chinese A stock market related companies’ financial and transaction data from December 2007 to December 2012. And finally, we get 9 efficiency factors which can be used to build our multiple-factor stock selection model.
Thirdly, we use the scoring method, principal component analysis, regression and K-Means clustering to build 6 strategies. Besides, we test back using these strategies based on data from January 2015 and December 2017. The research results show that apart from the principal component analysis method, the other five multiple-factor strategies performance well and the top group significantly outperforms the HS300 index. Taking annual return, sharpe ratio, excess odds and maximum drawdown into consideration, the clustering equal-weighted grading method performs best. Annual excess return reaches 23.515% and the excess winning percentage is greater than 60%.
However, there are still some shortcomings in our research. For example, there are several large drawdowns in back test and the depth and breadth of factor mining are not sufficient, which will be the main emphasis in our further study.
KEY WORDS: multiple-factor model, K-Means clustering, regression, principal component analysis
目录
摘要 I
Abstract II
第一章 绪论 1
1.1研究背景及意义 1
1.2文献综述 1
1.2.1国外论文综述 1
1.2.2国内文献综述 2
1.3研究思路与技术路线 4
1.4创新 5
第二章 相关理论概述 6
2.1 Markowitz投资组合选择理论 6
2.2资本资产定价模型(CAPM) 6
2.3套利定价理论(APT) 7
2.4 Fama-French三因子模型 7
2.5多因子选股模型介绍 8
2.5.1主成分分析 8
2.5.2 K-Means聚类 9
2.5.3线性回归 9
第三章 因子检验 10
3.1数据选取和预处理 10
3.1.1数据选取 10
3.1.2财务数据滞后处理 10
3.1.3数据频率转换 11
3.2因子选取与处理 11
3.2.1因子选取 11
3.2.2因子构建 11
3.2.3因子处理 14
3.3因子有效性检验 15
3.3.1回归法 15
3.3.2因子IC值 16
3.3.3排序打分法 16
3.4因子冗余性检验 17
第四章 多因子选股模型的构建 19
4.1换仓周期确定 19
4.2选股 19
4.2.1打分法 19
4.2.2主成分分析 19
4.2.3回归分析 20
4.2.4 K-Means聚类 20
4.3测试方法 22
4.4业绩评价 23
4.5实证结果分析 23
4.5.1样本内测试结果 24
4.5.2样本外测试结果 26
第五章 结论 33
5.1本文研究成果 33
5.2不足与改进之处 33
致谢 35
参考文献 36
附录 37
第一章 绪论
1.1研究背景及意义
量化投资,即用数量化的方法进行投资,运用统计、优化等数学方法建立模型,并用计算机语言进行实现。量化投资通过计算机程序来实现,因此可以将历史几十年地数据在几分甚至几秒回测出来,快捷而高效,并且受人类主观因素的影响较小。
剩余内容已隐藏,请支付后下载全文,论文总字数:45930字
该课题毕业论文、开题报告、外文翻译、程序设计、图纸设计等资料可联系客服协助查找;