【摘要】：在我國,股票市場已成為重要的投資渠道。投資者選擇不同的投資組合將面臨不同的風險,風險補償因此成為重要的定價因素,尋找有效的股票收益定價因子成為金融研究的熱點和難點。本文提出了一種新的因子構建方法,以提高模型定價水平。具體來說,本文認為使用單一異象變量進行因子構建,會使其包含較多噪聲,進而影響模型定價水平�；谶@一分析,本文利用三個可以反映盈利能力的指標,構建得到綜合盈利因子SPF;利用三個可以反映投資大小的指標,構建得到綜合投資因子SIF,據(jù)此得到一種四因子模型SCM。利用該四因子模型對賬面市值比因子HML和動量因子UMD進行定價,發(fā)現(xiàn)當考慮模型中四個因子之后,HML和UMD不再包含額外的定價信息。將SPF對市值因子MKT、規(guī)模因子SMB、HML、UMD和SIF進行回歸,剔除因子共有信息之后,得到一種新的綜合盈利因子SPFN。據(jù)此,本文構建了另一種四因子模型SCMN�；诶碚摲治�,可以預期這兩個因子模型的表現(xiàn)優(yōu)于本文涉及的其它5個模型。在實證檢驗部分,本文首先對MKT、SMB、HML、UMD、盈利因子RMW、投資因子CMA、SPF和SIF這8個因子的定價能力進行了比較分析。結果表明,MKT表現(xiàn)最優(yōu),SMB和SPF次之,這表明本文構建得到的綜合盈利因子優(yōu)于RMW。接著,本文利用隨機抽樣的方法進行了組合構建,使得組合不包含任何先驗信息,結果表明僅有MKT被選入模型。這一結果表明當利用異象變量進行組合構建時,其會包含先驗信息,會對因子有所偏好,進而使得模型檢驗結果存在偏誤。在模型檢驗部分,利用本文構建的18個檢驗組合對包含SCM和SCMN在內(nèi)的7個定價模型進行了檢驗,結果表明本文構建的SCMN和SCM在表現(xiàn)最優(yōu)次數(shù)和不被拒絕次數(shù)兩個層面均優(yōu)于其它5個模型。最后本文利用三種方法進行了穩(wěn)健性分析,以實現(xiàn)兩個目標:檢驗組合包含更多的先驗信息,同時獲得充分多的檢驗組合。檢驗結果表明,從表現(xiàn)最優(yōu)數(shù)量來看,SCMN最多,SCM次之;從模型不被拒絕比例來看,SCMN最高;從模型穩(wěn)定性來看,SCMN表現(xiàn)最穩(wěn)定,SCM次之。利用上述穩(wěn)健性分析,進一步證實本文構建的兩個模型有更高的定價能力。這一結果表明本文提出的因子構建方法能夠提高模型定價水平,而利用美國數(shù)據(jù)進行的實證檢驗同樣證實了這一點�；诶碚摲治龊蛯嵶C研究,并結合最新文獻成果,本文認為可以在如下兩個方面對因子模型進行更為深入的研究:一,如何在定價模型中包含更多信息,進而能對足夠多的異象變量進行解釋,這一點是很有研究價值的。是否能夠提出新的方法,使得定價模型在具有良好計量性質(zhì)的前提下,包含更多的定價信息,值得進一步深入研究;二,對非嵌套因子模型的差異顯著性進行分析研究,進而對定價模型含義有更深入的認知。
[Abstract]:In our country, the stock market has become an important investment channel. Investors will face different risks when they choose different portfolios, so risk compensation has become an important pricing factor. Finding an effective pricing factor of stock returns has become a hot and difficult point in financial research. In this paper, a new method of factor construction is proposed to improve the pricing level of the model. Specifically, this paper argues that the use of a single aberrant variable for factor construction will make it contain more noise, thus affecting the pricing level of the model. Based on this analysis, this paper constructs a comprehensive profit factor SPF; by using three indexes that can reflect profitability. Using three indexes which can reflect the investment size, we construct the comprehensive investment factor SIF, and get a four-factor model SCM.. The four-factor model is used to price the book-to-market ratio factor (HML) and momentum factor (UMD). It is found that HML and UMD no longer contain additional pricing information after considering the four factors in the model. The SPF regression is applied to the market value factor MKT, scale factor SMB,HML,UMD and SIF. After removing the common information of the factors, a new comprehensive profit factor SPFN. is obtained. Based on this, another four-factor model, SCMN., is constructed in this paper. Based on the theoretical analysis, it can be expected that the performance of these two models is better than the other five models. In the part of empirical test, this paper firstly analyzes the pricing ability of MKT,SMB,HML,UMD, profit factor, RMW, investment factor CMA,SPF and SIF. The results show that MKT is the best, SMB and SPF are the second, which indicates that the comprehensive profit factor constructed in this paper is superior to RMW.. Then, the method of random sampling is used to construct the combination so that the combination does not contain any prior information. The results show that only MKT is selected into the model. The results show that when the visionary variables are combined, they will contain prior information and have a preference for the factors, which will make the model test results biased. In the part of model checking, seven pricing models, including SCM and SCMN, are tested by using the 18 test combinations constructed in this paper. The results show that the SCMN and SCM constructed in this paper are superior to the other five models in terms of the optimal times of performance and the times of non-rejection. Finally, the robustness analysis is carried out by using three methods to achieve two objectives: the test combination contains more prior information and the sufficient number of test combinations is obtained at the same time. The test results show that, in terms of the optimal number of performance, SCMN is the most, SCM is the second, SCMN is the highest in terms of the proportion of model not rejected, SCMN is the most stable in terms of model stability, and SCM is the second. By using the above robust analysis, it is further proved that the two models constructed in this paper have higher pricing power. The results show that the proposed method of factor construction can improve the pricing level of the model, and the empirical test using American data also confirms this point. Based on the theoretical analysis and empirical research, combined with the latest literature, this paper thinks that we can do more in-depth research on the factor model in the following two aspects: first, how to include more information in the pricing model, It is very valuable to explain enough aberration variables. Whether we can put forward a new method to make the pricing model contain more pricing information under the premise of good metrological property is worthy of further study; Secondly, the significance of the non-nested factor model is analyzed, and the meaning of the pricing model is further recognized.
【學位授予單位】：南京大學
【學位級別】：碩士
【學位授予年份】：2017
【分類號】：F224;F832.51
，

本文編號：2415751