基于公因子提取法的股票定价有效性研究

发布时间：2018-01-03 12:41

本文关键词：基于公因子提取法的股票定价有效性研究　出处：《浙江工商大学》2014年硕士论文　论文类型：学位论文

【摘要】：Sharpe (1964)提出的资本资产定价模型(CAPM)是研究金融市场定价理论的里程碑,认为在资产定价过程中,只有系统风险起着不可替代的作用,而非系统风险可以通过分散化投资得以消除。然而受不完全信息、交易成本等因素的影响,投资者未能持有多样化的投资组合,非系统风险固然存在,导致CAPM这一单因素模型的定价功能逐渐减弱。随之,多变量资产定价模型成为现代金融界研究的热点之一。然而随着对定价模型研究的发展和深入,越来越多的因素被纳入到定价研究的范畴来解释各种截面异象。实证中由于对这么多因素做回归可能存在多重共线性而导致得不到真实的结果。本文尝试基于渐近主成分分析的高维因子分析法,其核心思想是将高维时间序列转换有几个时间序列组成的低维时间序列,也即公因子序列,这几只公因子序列反映了数据的绝大部分变动。然而,实际应用中这些公因子往往是未知的,也叫潜因子,公因子的个数也是未知的需要估计。本文基于主成分分析方法估计出公因子及其个数后,将已知或可观测到的因素与少量的潜因子建立线性回归模型,通过一系列的统计量来判定真实的潜因子是否能表示实证因子(即可观测因子),从而检验实证因子的有效性。公因子提取方法的引入,为检验众多名目繁多的实证因子的有效性提供了思路。本文以A股市场为研究对象(共计796只股票,样本期间自2009年2月至2013年12月),首次从潜因子识别的角度对可观测变量进行有效性检验。主要结论如下：第一,在潜因子估计阶段,通过Bai和Ng(2002)所建立的面板准则对超额收益率序列进行潜因子个数的相合估计,将维度高达796的高维时间序列转换为3维时间序列。第二,在对可观测变量的构造过程中发现,中国A股在样本期内市场不存在规模效应,上市公司股票的平均月收益率随着公司市值的上升而上升。在动量因子的研究中发现中国A股市场存在短中期动量效应,长期反转效应。第三,在可观测变量与潜因子的回归模型估计结果中,发现假设残差具有条件异方差性条件下各个统计量估计结果要优于独立同分布假设条件下的结果。在对微观经济变量与潜因子的关系研究中,市场溢酬因子和账面市值比因子可以看成是潜因子的替代变量。在对宏观经济变量与潜因子的关系研究中,发现宏观变量都不及微观变量对股票收益率的变动影响,但相对来说,消费者满意程度和通货膨胀率与潜因子的相关性最强。第四,在分行业研究各可观测变量对不同行业股票收益率的影响程度中,市场溢酬因子对各行业股票收益率变动的影响显著,宏观经济变量中的通货膨胀率则对股票定价起了不可忽视的作用。
[Abstract]:Sharpe (1964) put forward the capital asset pricing model (CAPM) is a milepost on financial market pricing theory, in the process of asset pricing, only plays an irreplaceable role in the system risk and non system risk can be eliminated through portfolio investment. However due to incomplete information, transaction costs and other factors. Investors failed to hold a diversified portfolio of non system risk is exist, resulting in the pricing function model of CAPM the single factor decreases gradually. Then, the multi variable asset pricing model become a hot research topic in modern financial circles.
However, with the development of research on the pricing model and in-depth, more and more factors are incorporated into the scope of the study of pricing to explain various empirical section vision. Because of so many factors may do regression multicollinearity caused no real results. High dimensional factor this paper attempts based on the asymptotic analysis of principal component analysis method, its core idea is the high dimension time series conversion of low dimensional time series is composed of several time series, namely the common factor sequence, these factors reflect the sequence of most data changes. However, the practical application of these factors is often unknown, also called latent factors, need factors the number is unknown. Estimates based on principal component analysis method to estimate the common factor and the number after the factors known or observable and a latent factor to establish the linear regression model, by A series of statistics is used to determine whether the real latent factor can represent the empirical factor (the observation factor), so as to test the effectiveness of the empirical factor. The introduction of the common factor method provides a train of thought for testing the effectiveness of many various empirical factors.
In this paper, the A stock market as the research object (a total of 796 stocks, the sample period from February 2009 to December 2013), for the first time from the perspective of identifying latent factor to test the effectiveness of observable variables. The main conclusions are as follows: first, the latent factor estimation stage, by Bai and Ng (2002) Consistentestimate panel established guidelines a number of potential factors on excess return series, high dimensional time series will be as high as 796 dimensions into 3 dimensional time series. Second, found in the construction process of observable variables, China A shares in the sample period does not exist in the market, shares of listed companies the average monthly rate of return to rise with the market value of the company. In the study of the momentum factor found in the China A-share market there is A short term momentum effect, long-term reversal effect. Third, the regression model of observable variables and latent factor estimation result in false A residual with conditional heteroscedasticity under each statistic estimation results is better than the i.i.d. assumption results. In the study on the relationship between the micro economic variables and latent factor, market factor premium and book to market factor can be regarded as a substitute for latent factor variables. In the research on the relationship between macroeconomic variables and the latent factor, found that the influence of the macro variables are not macro variables on the stock changes in the rate of return, but relatively speaking, the strongest correlation between customer satisfaction and the rate of inflation and the latent factor. In fourth, the industry research variable degree of influence on the stock returns in different industries, the market premium of significant factor the industry rate of stock returns and macroeconomic variables in the inflation rate on stock pricing plays a role can not be ignored.

【学位授予单位】：浙江工商大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：F832.51;F224

【参考文献】