1
$\begingroup$

I am doing a Engle-Granger test for cointegration and I am unsure about some commands.

"Cointegration and the ECM" (document) from learneconometric.com says I should use:

regress b f
predict ehat, residual
regress D.ehat L.ehat L.D.ehat, noconstant

However, "Time series" (document from Princeton Uni) says I should use:

regress b f
predict ehat, resid
dfuller ehat, lags(10)

So I am unsure about the last commands here. Should I use regress D.ehat L.ehat L.D.ehat, noconstant or dfuller ehat, lags(10) and what is the difference here? Also, how many "lags" should I include for the Dickey Fuller test?

$\endgroup$
1
$\begingroup$

Do the last one, the first one is just the same thing but you will not be using the in-built adf function. The second one does it better and you have a choice of including lagged differences to control for possible autocorrelation. If your data is monthly, give it a lag order of 12.

| improve this answer | |
$\endgroup$
0
$\begingroup$

In STATA I would actually recommend the package egranger

Find more about it:

findit egranger

and to install it:

ssc inst egranger

Regarding the DF-test, I usually use another package, dfao. This one lets me use an information criterion, such as HQIC, AIC or SBIC, to choose the optimal lag length. Again:

findit dfao

To install it:

ssc inst dfao

My default specification for the DF is

dfao varname, lltc(sc) noao notr

where the options specify, in order: use the Schwarz criterion to choose the lag length; do not test for additive outliers [i.e. work like normal DF test]; and do not add a trend automatically.

| improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.