PANELDOLS—Pedroni(2001) DOLS

[TomDoan]

This is a revised procedure for estimating the cointegrating vectors using the multivariate group mean panel DOLS from Pedroni(2001) "Purchasing Power Parity Tests in Cointegrated Panels," Review of Economics and Statistics, 83, 727-731.

Compared with earlier versions, this adds the option for DET=TREND (not just CONSTANT), a SMPL option for excluding data points, defines global variables for the individual coefficients and covariance matrices and defines residuals from the mean group estimates.

paneldols.src

Detailed description

The use of this procedure is covered in detail as part of the Panel/Grouped Data e-course.

A companion procedure for estimating with FM OLS rather than DOLS is @PANELFM.

An example of its use is Pedroni(2001)

Important Note

DOLS can very quickly exhaust the degrees of freedom in a data set. There are 2*DLAGS+1 (by default 5) added regressors in the differences for each right side endogenous variable plus you lose DLAGS*2+1 data points allowing for lags and leads and differences. So 20 observations per individual, two right side endogenous variables, DLAGS=2 leaves you with 15 usable observations, and 13 regressors (constant + 2 current RHS + 2 x 5 additional lags and leads on the differences).

[sanjeev]

Dear Tom,
Can you please tell me the option to set the number of lags and leads in case of group-mean DOLS estimation in a panel data?


Thanks,
Regards.

[TomDoan]

Click on the detailed description---the option you're looking for is DLAGS.

[sanjeev]

Dear Tom,
Can we specify the number of lags separately from number of leads in group-mean panel DOLS estimation?

Regards.

[TomDoan]

No. Why would you want to do that? Those aren't part of the model---they're incidental regressors added to soak up the simultaneity.

[sanjeev]

Dear Tom,
Actually when I use group-mean panel DOLS procedure on my data(1992-2013) for eight members,I get t-statistics to be very large numbers(in thousands) using default numbers of two leads and 2 lags. Even if I specify dlags=1,I get the same kind of results.Is it possible to specify different umber of lags and leads in this estimation procedure?

Please reply soon!


Thanks,
Regards.

[TomDoan]

I'm confused. Why is that a problem? And why do you think changing up the leads/lags will "fix" it? The fact that the results are robust to changes in the value of DLAGS is a good sign, not a bad one since DLAGS covers "nuisance" parameters.

[sanjeev]

Dear Tom,
Thank you for your quick reply! Actually,the problem is that such high t-statistics(in thousands) don't make sense to me.Secondly,when I use zero lags and leads,then the t-statistics look normal(not such high values).Please suggest what should I do?


Regards.

[sanjeev]

This is in continuation with my previous mail.
Another problem by using group-mean panel DOLS procedure that I am facing is that even the coefficients are also unexpectedly high,not making any sense.

Regards.

[TomDoan]

It sounds like you have either bad data or a bad model. Overly large and very significant coefficients sounds like you're estimating something other than what you actually want.

[sanjeev]

Dear Tom,
Does it make any sense to estimate a group mean panel DOLS model using zero dlags?
Secondly, where can I find the program for doing panel DOLS,the pooled version?

Thanks,

Regards.

[TomDoan]

Re DLAGS=0, probably not. It would require there to be effectively no short-run dynamics in the model at all.

Isn't there a link to an example above?