论文部分内容阅读
从统计意义上比较不同模型的改进效力有助于挑选出最接近金融数据生成过程的定价模型,是资产定价研究的重要课题。我们借鉴Kan和Robotti~([1])的研究成果,基于第一HJ距离构造了广义似然比检验,以台湾市场丰富的数据资料为基础,对8种常见的线性因子模型(包括基于金融资产价格的线性因子模型)进行了模型两两差异性检验。研究发现:VM和CAPM、FF3和LM这两组模型无明显差异,表明波动率冲击因子和流动性因子未带来显著的模型改进效力。由于部分定价因子可能具有共同的解释能力,VM和IVM、IVM和HSM、HSM和VanM、VanM和SkewM这多组模型间也未表现出显著的差异。同时,引入条件信息是否能改善模型效力视不同模型而定,在10%的显著性水平下,FF3、LM、VanM、SkewM的条件信息模型较无条件信息模型有所改进。
Comparing the improved effectiveness of different models in a statistical sense helps to pick out the pricing model that is closest to the financial data generation process and is an important issue in asset pricing research. Based on the research results of Kan and Robotti ~ ([1]), we construct the generalized likelihood ratio test based on the first HJ distance. Based on the abundant data in the Taiwanese market, we analyze eight common linear factor models (including those based on the financial Asset price linear factor model) conducted a two-for-two difference test. The study found no significant difference between VM and CAPM, FF3 and LM models, indicating that volatility shock factors and fluidity factors did not bring significant model improvement efficacy. Since some pricing factors may have common interpretive power, no significant differences were found between the multiple models of VM and IVM, IVM and HSM, HSM and VanM, VanM and SkewM. At the same time, whether the introduction of conditional information can improve the effectiveness of the model depends on different models. Under the 10% significance level, the conditional information model of FF3, LM, VanM and SkewM is improved compared with the unconditional information model.