具有内生性与时变方差的线性模型参数稳定性检验

Stability Test on Linear Model Parameters With Endogeneity and Time-varying Variance

时间序列数据往往同时存在内生性和时变非线性特征,从而使得计量模型的检验变得更加复杂。文章基于矩条件稳定性检验的U-统计量,提出一种诊断回归模型参数稳定性的新方法。该方法同时适用于内生性与时变方差情形,可以检验各种形式的结构变化,并在原假设下服从渐近分布,在对立假设下也具有较高的检验功效。将该方法应用于检验我国新凯恩斯混合Phillips曲线,发现其存在明显的结构变化。

Time series data are often characterized by both endogeneity and time-varying nonlinearity, which makes the testing on econometric models more complicated. This paper is based on the U-statistics of moment condition stability test to propose a new method of diagnosing the parameter stability of regression models. The proposed method is suitable for both endogeneity and time-varying variance, and can test all kinds of structural changes. It follows asymptotic distribution under the null hypothesis,and has high test efficiency under the opposite hypothesis. Finally, the paper applies this method to test the New Keynes Mixed Phillips curves in China, finding that there exist obvious structural changes.