摘要
近年来，深度学习技术被应用于资产定价模型的修正过程中，并取得了良好的
效果。但是此类研究基本都是针对国外证券市场进行的，将深度学习应用到国内 A
股市场进行资产定价研究的文章相对匮乏。本文基于 Feng et al.(2018)提出的深度学
习框架，首次结合中国 A股市场上市公司的基本面数据，搭建了一个可以解释和预
测中国 A股市场投资组合收益率的深度因子模型。
本文选取 2003年 1月至 2020年 8月共 212个月的上市公司月度特征数据和相
应的宏观经济政策指标数据，利用深度学习模型框架从非线性因素、时变因素、宏
观经济因素等多个角度对 A股市场中的资产定价模型进行系统的研究。实证结果显
示：(1)加入深度因子确实可以提升原有模型的解释力； (2)浅层的神经网络构建的
深度因子模型比深层的神经网络表现更优；(3)时变因子模型并不总是优于非时变因
子模型；(4)加入宏观经济因素的模型对于 A股市场的解释力更强。
本文还利用深度学习技术进一步比较了各个公司特征在模型解释能力中的贡献
度，发现换手率对于 A股市场的定价模型来说是一个较为重要的指标。本文的研究
对如何将深度学习应用于我国 A股市场中的资产定价领域有重要的引导意义。
关键词：资产定价；深度因子模型；深度学习；模型修正
I

Abstract
In recent years, deep learning technology has been applied to the revision process of
asset pricing models, and it has achieved good results. However, this kind of research is
basically carried out on foreign securities markets. The paper about applying deep
learning to the domestic A-share market for asset pricing research are relatively scarce.
This paper bases on the deep learning framework proposed by Feng et al. (2018) and
combines it with the firm characteristics in China A-share market. This paper builds a
deep factor model that can explain and predict the return rate of investment portfolio in
China A-share market.
This paper chooses 212 months, from January 2003 to August 2020, of data which
provided by those listed companies and the corresponding macroeconomic policy
indicator as the research database. By using the deep learning model framework to
analyze these data from multiple perspectives such as non-linear factors, time-varying
factors, and macroeconomic factors, we can systematically study asset pricing models in
the A-share market. The empirical results show that: (1) adding the deep factor can
improve the explanatory power of the original model; (2) the deep factor model
constructed by the shallow neural network performs better than the deep neural network;
(3) the time-varying factor model is not always better than the non-time-varying factor
model; (4) the model with macroeconomic factors has a stronger explanatory power for
the A-share market.
This paper uses deep learning technology to further compare the contribution of each
firm characteristics to the model interpretation ability. The result shows that the turnover
rate is a very important indicator for asset the price in China A-share market. The research
in this paper has important guiding significance for how to apply deep learning to asset
pricing in the A-share market.
Keywords: Asset pricing; Deep factor model; Deep learning; Model modification
II

目录
摘要............................................................................Ⅰ
Abstract......................................................................... II
目录........................................................................... III
1
绪论.......................................................................... 1
研究背景.....................................................................1
研究意义.....................................................................4
研究内容及方法...............................................................5
技术路线图...................................................................6
创新与贡献...................................................................7
文献综述...................................................................... 8
资产定价的线性模型...........................................................8
资产定价的机器学习模型......................................................10
文献评述....................................................................12
方法介绍..................................................................... 14
理论基础................................................................... 14
深度学习网络................................................................15
深度学习模型的目标函数......................................................16
深度学习实现 Fama-French模型................................................17
深度学习因子模型构建......................................................... 21
非线性特征..................................................................22
深度因子模型................................................................24
时变深度因子模型............................................................26
整个模型的框架结构..........................................................27
实证研究..................................................................... 29
样本选择....................................................................29
模型评估指标................................................................31
实证过程....................................................................32
总结......................................................................... 44
研究结论....................................................................44
启示与建议..................................................................45
1.1
1.2
1.3
1.4
1.5
2
2.1
2.2
2.3
3
3.1
3.2
3.3
3.4
4
4.1
4.2
4.3
4.4
5
5.1
5.2
5.3
6
6.1
6.2
参考文献........................................................................ 47
致谢............................................................................ 51
III
。。。以下略