Re: [Gretl-users] DF unit root test for residual process in cointegration equation
On Sat, 27 Sep 2008, Piotr Kowalczyk wrote:
I have just subscribed to Gretl users group. So firstly, I would
like to welcome all users and apologize for all my language
mistakes.
Welcome!
I have got a problem with my analysis on cointegration... I was
wondering if someone could send me the proper way to test
cointegration without using Gretl so as to achieve the same
results as using Gretl. I am interested in no constant test with
0 and 1 lag. Where I can find the information on how Gretl is
estimating tau_nc, estimated (a-1) and p-value for this test.
Here's an example for the 0 lag case.
<script>
open data9-7
# Engle-Granger
coint 0 UNEMP PRIME --nc
# Manual reconstruction of test regressions
diff UNEMP PRIME
ols d_UNEMP UNEMP(-1) --simple-print
ols d_PRIME PRIME(-1) --simple-print
ols UNEMP PRIME --simple-print
genr u = $uhat
diff u
ols d_u u(-1) --simple-print
</script>
You should find that the "ols" commands produce the same output as
the regressions reported by "coint" (Engle-Granger test), apart
from the p-values. The standard p-values reported in the context
of OLS are not applicable in the Dickey-Fuller context, and the
values reported in connection with the coint command are based on
J. MacKinnon's work. See "help coint", "help adf".
Allin Cottrell