基于第一HJ距离的线性因子模型两两比较研究

Pairwise Comparisons of Linear Factor Models Based on The First HJ Distance

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。

Performance Comparisons between different models statistically contributes to pick out the pricing model that is most proximal to the true generating process of financial data, and it＇s important for asset pricing research. Referring to Kan and Robotti, we use the first HJ distance to construct the general likelihood ration test, and test the pairwise models＇ performance differences of eight linear factors models（including models based on financial asset prices）using the Taiwan market data and find that： VM and CAPM, FF3 and LM show no difference, impliing that the stock market volatility shock and liquidity risk brings in no significant model improvement. Since parts of factors have the common explanation ability, VM and IVM, IVM and HSM, HSM and VanM, VanM and SkewM these groups also exhibite performance indifference. Meanwhile, whether conditional information improves model performance or not varies with different models. Under the 10% significance level, the conditional versions of FF3, LM, VanM, SkewM improves statistically compared with the unconditional ones.