Unit root test (ADF)

[johnvaz@gmail.com]

I am getting contradictory prob values for the same critical values of tau from the ADF test and the ADF test embedded in the engel granger coint test. Note the same critical value is reporting different prob* values. There are two tables reported below the Uroot test and then coint.
Is this a bug?

ADF test:

Null Hypothesis: ZLMBROAD_RES has a unit root
Exogenous: None
Lag Length: 0 (Automatic - based on SIC, maxlag=14)

t-Statistic Prob.*

Augmented Dickey-Fuller test statistic -2.919045 0.0036
*********************
Test critical values: 1% level -2.576693
5% level -1.942439
10% level -1.615633

*MacKinnon (1996) one-sided p-values.

coint test (engel and granger)

Series: LS LMBROAD
Sample: 1990M09 2007M03
Included observations: 199
Null hypothesis: Series are not cointegrated
Cointegrating equation deterministics: C
Automatic lags specification based on Akaike criterion (maxlag=14)


Dependent tau-statistic Prob.* z-statistic Prob.*
LS -2.919045 0.1355 -15.02584 0.1400
**********************************
LMBROAD -2.784998 0.1751 -13.86581 0.1757

*MacKinnon (1996) p-values.

[EViews Glenn]

This is not a bug. You should get a different probability value. The standalone ADF test on the residuals doesn't know that you are computing the test on the residuals from the assumed cointegrating regression. The Engle-Granger version of the test explicitly accounts for this.

If you look at discussions of the Engle-Granger test in the literature or in textbooks you will often see explicit mention of the fact that while you can construct the test statistic by simply computing the ADF on the residuals, the actual critical values and probabilities must be adjusted to account for the initial estimation.