Improved critical values for the Enders-Granger unit-root test

Abstract

Enders and Granger provide critical values to test the null hypothesis of a unit-root against the alternative of threshold adjustment. However, in obtaining their critical values, Enders and Granger did not use a consistent estimate of the threshold nor did they use a lag-augmented data generating process. This note remedies both of these problems. The power of the test statistics using the consistent estimate of the threshold are compared to those of Enders-Granger and of Dickey-Fuller. Surprisingly, the original Enders-Granger statistic often has the highest power. As such, the Enders-Granger statistic using a lag-augmented data generating process is calculated.