STEP1: as show in figure A2.

1, start with the least restrictive of the plausible models (which will
generally include a trend and drift) and use the "t" statistic to test the null hypothesis
lambda=0. Thus, in the most general case, you estimate the model in the form of (4.25) so
that...

Unit root tests have low power to reject the null hypothesis; hence, if the null hypothesis of a
unit root is rejected. there is no need to proceed. Conclude that the "y" sequence does not
contain a unit root.

STEP 2: If the null hypothesis is not rejected, the next step is to determine whether the trend
belongs in the estimating equation. Toward this end, you test the null hypothesis a2=lambda=0
using phi3 statistic. If you do not reject the null hypothesis, assume the absence of a trend and
proceed to step3.

If you have reached this point, it is because the "t" test indicates that there is a unit root, and
the phi3 test indicates that lambda and/or a2 differs from zero.

As such, it would seem that there is a unit root and a trend. You can gain additional support for
this result by assuming that there is a unit root and estimating "diferenciay=...". SInce there are
no I(1) regressors in this specification, the test for the null hypothesis a2=0 can be conducted
using a standard "t" test. If you conclude a2=0, go to step 3 since it does not appear that there
is a trend. If you find that "a2=/0", use (4.25) to test the null hypothesis lambda=0 using a t-
distribution. Given that the trend belongs in the regression equation, rule 2 indicates that the
test for lambda=0 can be conducted using a t-distribution. Go no further, if lambda=/0,
conclude that the sequence is trend-stationary. If lambda=0, conclude that there is a unit root
(and the "y" sequence contains a quadratic trend)

STEP 3: Estimate the model without the trend (i.e., estimate a model in the form of (4.24)).
Test for the presence of a unit root using the "tu" statistic. If the null hypothesis is rejected,
conclude that the model does not contain a unit root. If the null hypothesis of a unit root is not
rejected, test for the significance of the intercept by testing the hypothesis a0=lambda=0 using
the phi1 statistic. If you do not reject the null hypothesis a0=lambda=0, assume that the
intercept is zero and proceed to step 4. Otherwise, estimate "y" and test whether a0=0 using a
t-distribution. If you find a0=0, proceed to step 4. Otherwise, conclude that process contains
an intercept term. In accord with rule 2, you can use a t-distribution to test whether lambda=0
in the regression "y". If the null hypothesis of a unit root is rejected, conclude that the "yt"
sequence is stationary around a nonzero mean. Otherwise, conclude that the "yt" sequence
contains a unit root and a drift.

STEP 4: Estimate a model without the trend or drift; i.e., estimate a model in the form (4.23).
Use "t" to test for the presence of a unit root. If the null hypothesis of a unit root is rejected,
conclude that the "yt" sequence does not contain a unit root. Otherwise, conclude that the
"yt" sequence contains a unit root.