时序非平稳性ADF检验法的理论与应用

ADF test for non-stationary time series:theory and application

数据非平稳问题是经典和非经典计量经济学的分水岭,在经典计量经济学领域内,假设变量是平稳的,因此不考虑变量的非平稳性,而实际上大多数经济数据又是非平稳的.研究表明,用非平稳变量进行回归分析降低了检验的功效,结果常常是虚假回归,所以对变量进行平稳性检验是避免虚假回归的前提.参数检验的ADF法作为常用的数据非平稳性的检验方法,有其自身的理论发展脉络和应用价值.通过对香港向内地输入的影响因素的分析,发现香港对内地输入与其影响因素——中国的GDP以及港币汇率均是I(1)变量,且三者具有协整关系而且是非线性函数关系.随着面板数据使用的增加,ADF检验法将逐渐更广泛地应用于面板数据的分析和检验过程中.

The problem of the non-stationary data is the watersheds of classics and non-classical econometrics. In the domain of classical econometrics, the variable is supposed to be stationary, so stationarity of variables is not under consideration, but actually most economic data are non-stationary. Studies show that using non-sta- tionary variables for regression analysis would influence the effectiveness of the test and would easily come to dummy regression. Therefore, in view of avoiding dummy regression, a test for stationarity of variables is neces- sary. The parameter estimation-ADF is often used for this test and has its own theory development and value. Through the analysis of export from Hong Kong to China-mainland, we find out the factors which affect the ex- port. The GDP of China and the currency rate is Ⅰ （ 1 ） variable, and there exists co-integration and non-linear function between these 3 variables. With the increase of utilizing of panel-data, the ADF will be used extensively in the analysis and test of panel-data.