Augmented Dickey-Fuller Test and the Lag Length Selection Problem

Many studies indicated that ADF test is very sensitive to different leg length selection models. Based on Hall, and Ng, Perron’s works, this article simulates a more general ARIMA(0,1,q) process and compares the influence of different selection methods to the size and power of the ADF test. Finally, it is proved that the Modified Information Criteria always shows a more proper size and the General to Special Criteria has more robust ADF test properties.