基于分位数回归的特质风险溢价研究

发布时间：2018-04-29 21:08

本文选题：特质风险 + 分位数回归　；参考：《浙江工商大学》2012年硕士论文

【摘要】：根据经典的资本资产定价理论,特质风险作为非系统风险可通过投资组合进行分散。Merton(1987)基于信息不对称理论认为：投资者只对自己熟悉的投资机会进行投资选择,公司特质风险难以被分散。特质风险的定价问题也随之成为当前金融领域研究的热点之一。然而,实证结论的不一致导致该问题争议颇多。这些争议可能是由定价模型本身的假定所引起,也可能是检验工具的不同所致。本文尝试从定价模型选取和检验方法的改进两方面对特质风险的定价问题进行探讨。具体而言,定价模型选取方面,定价模型本身的设定问题可能会影响检验结果。Fama-French三因子模型模型解释了金融市场中不能被传统资产定价模型所解释的大量金融异象,并且成为特质风险定价研究的通用基准依据。然而定价模型不同,特质波动率是否会产生大的差异?本文尝试采用Carhart四因子模型来计算特质波动率,衡量特质风险。并比较基于不同定价模型,计算得到的特质波动率的差异。另外,在检验特质风险定价情况的过程中,笔者认为常用的均值回归模型难以全面衡量特质风险的溢价情况。本文首次尝试采用分位数回归的方法研究对于不同收益率部分特质风险与横截面收益的关系,探讨对于不同收益率部分特质风险的定价是否存在显著差别。本文以A股市场为研究对象,从以上两个新的视角检验中国股票市场特质风险是否被定价。主要结论如下：1.在特质风险的估算阶段,动量因子的加入使特质风险的估计结果略小于采用FF.3模型的估计结果。这说明信息传递速度对特质风险的估算有一定的影响。2.根据不同分位点的回归结果发现,与均值回归模型估计结果不同,特质风险并不是简单的折价或者溢价。特质风险不仅伴随时变性,而且对于不同收益率部分特质风险的定价情况存在显著差别。随着分位点的变化,特质风险与横截面收益的回归系数出现了在低分位点不显著或显著为负的情况；而在高分位点,特质风险在1%的显著性水平下存在溢价情况。这说明,在低收益率部分特质风险的定价情况不稳定,特质风险溢价、折价情况都有存在：在高收益率部分特质风险不存在异常收益之谜。这也表明,相较单一的均值回归模型,分位数回归的方法在解释特质风险定价情况时更为全面。另外本文分析发现规模、账面市值比、换手率、流动性、反转效应等公司特征变量对特质风险的定价问题没有显著性影响。
[Abstract]:According to the classical capital asset pricing theory, idiosyncratic risk as a non-systematic risk can be dispersed by portfolio. Merton 1987. based on the information asymmetry theory, it is believed that investors only make investment choices to their familiar investment opportunities. Corporate trait risk is difficult to spread out. The pricing of idiosyncratic risk has become one of the hotspots in the field of finance. However, the inconsistency of empirical conclusions leads to a lot of controversy on this issue. These disputes may arise from the assumptions of the pricing model itself or from the differences in the testing tools. This paper attempts to discuss the pricing of idiosyncratic risk from two aspects: the selection of pricing model and the improvement of test methods. In particular, in the aspect of pricing model selection, the problem of pricing model itself may affect the test results. Fama-French three-factor model explains a large number of financial anomalies that cannot be explained by traditional asset pricing models in financial markets. And become the general benchmark basis for the study of idiosyncratic risk pricing. However, if the pricing model is different, will the volatility of traits vary greatly? This paper attempts to use the Carhart four-factor model to calculate the volatility of trait and measure the risk of idiosyncrasy. And compare the differences of idiosyncratic volatility based on different pricing models. In addition, in the process of testing the pricing of trait risk, the author thinks that the commonly used mean regression model is difficult to measure the premium of trait risk. This paper first attempts to use the quantile regression method to study the relationship between the cross-section return and the partial trait risk of different returns, and to explore whether there are significant differences in pricing the partial trait risk with different rates of return. In this paper, A-share market is taken as the research object to test whether idiosyncratic risk of Chinese stock market is priced from the above two new perspectives. The main conclusions are as follows: 1. In the stage of estimating trait risk, the addition of momentum factor makes the estimation result of trait risk slightly smaller than that of FF.3 model. This shows that the speed of information transfer has a certain impact on the estimation of trait risk. 2. According to the regression results of different loci, it is found that trait risk is not a simple discount or premium, which is different from that estimated by the mean regression model. Trait risk is not only accompanied by time-varying, but also has significant difference in pricing of partial trait risk with different yield. With the change of loci, the regression coefficient between trait risk and cross-sectional income was not significant or significantly negative at the low score, while at the high score, there was a premium at the significant level of 1%. This shows that in the low yield part of idiosyncratic risk pricing is unstable, the trait risk premium and discount exist: in the high yield part of the trait risk there is no abnormal return mystery. It also shows that the quantile regression method is more comprehensive in explaining the pricing of trait risk than the single mean regression model. In addition, this paper finds that the scale, book market value ratio, turnover rate, liquidity, reverse effect and other company characteristics have no significant impact on the pricing of idiosyncratic risk.
【学位授予单位】：浙江工商大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：F224;F830.91

【参考文献】