摘要
协整检验是进行回归分析的首要过程,是避免伪回归的主要方法.然而,大多数协整检验技术都是建立在非稳健的普通最小二乘框架下.这对于普遍具有尖峰厚尾的时间序列来说,可能会导致统计检验的失效.为了解决这个困难,本文提出带线性时间趋势模型的分位数回归协整检验方法.不同于传统的静态协整分析,我们构建了一个分位数残差累积和(QCS)统计量来检验不同分位点上变量间的动态协整关系.应用分位数回归和泛函极限理论,推导出了统计量的渐近分布,并提出了修正的QCS统计量,拓展了其在序列相关以及长期内生性模型中的应用.模拟给出了统计量的临界值并证明了本文的协整检验方法具有良好的有限样本性质.最后,利用所提方法,检验了可支配收入与实际消费之间的动态协整关系,发现随着分位点的增大,它们之间的协整关系越强.
Cointegration test is the primary process of regression analysis,and is the main way to avoid pseudo regression.However,most techniques of cointegration test are executed in the framework of non-robust ordinary least squares.This may lead to the failure of statistical tests for time series with peaks and thick tails.To solve this problem,the present article suggests the method of quantile cointegration test to investigate the cointegration model with linear time trend.Unlike the traditional static cointegration test,a quantile residual cumulative sum statistic(QCS)is constructed to investigate the dynamic cointegration relationship among variables at different quantiles.The asymptotic distribution of statistics is derived using quantile regression and functional limit theory.Furthermore,the modified QCS is proposed for application in sequence correlation and long-term endogenous models.The simulation results show that the cointegration test method presented in this paper has good finite sample properties.Finally,the cointegration linkage between disposable income and actual consumption is examined by the proposed method.
作者
解其昌
孙乾坤
XIE QICHANG;SUN QIANKUN(School of Finance,Shandong Technology and Business University,Yantai 264005,China)
出处
《应用数学学报》
CSCD
北大核心
2020年第3期555-571,共17页
Acta Mathematicae Applicatae Sinica
基金
教育部人文社科基金青年基金资助项目(16YJC790040).
关键词
协整检验
线性趋势
分位数回归
QCS统计量
cointegration test
linear trend
quantile regression
QCS statistics estimating equation
