Panel cointegration estimations - possible with Eviews 7?

For technical questions regarding estimation of single equations, systems, VARs, Factor analysis and State Space Models in EViews. General econometric questions and advice should go in the Econometric Discussions forum.

Moderators: EViews Gareth, EViews Moderator

maragloria
Posts: 101
Joined: Tue Jun 29, 2010 7:09 am

Panel cointegration estimations - possible with Eviews 7?

Postby maragloria » Fri Jan 21, 2011 7:00 am

Hi there,

I would like to know if the cointegration methods FM-OLS and CCR are available with panel structured data in Eviews 7.
If not, would you have any suggestions on how to do panel cointegration estimations? I already did the DOLS (leads & lags of Stock & Watson) manually and I would like to test for robustness with the other cointegrating methods.

Thanks,

Mara

EViews Glenn
EViews Developer
Posts: 2671
Joined: Wed Oct 15, 2008 9:17 am

Re: Panel cointegration estimations - possible with Eviews 7?

Postby EViews Glenn » Fri Jan 21, 2011 10:10 am

Non-panel FMOLS and DOLS are supported as built-in features in EViews 7 for non-panel structured workfiles. It is possible that you could use these new FMOLS, DOLS tools to, with a bit of work, construct a panel version of the those estimators (by averaging over the individual cross-section estimates), but this is not viewed as a standard EViews 7 feature. The non-averaging forms of the estimators would be more difficult to estimate.

This is on our list of things to look at for upcoming versions.

jayvesh
Posts: 1
Joined: Tue Jul 05, 2011 6:53 pm

Re: Panel cointegration estimations - possible with Eviews 7

Postby jayvesh » Tue Jul 05, 2011 6:58 pm

Hi,

This suggestion seems good. However, I would like to know whether the averaging can also be applied to the test stastics and critical values.

Thanking you in anticipation.

EViews Glenn
EViews Developer
Posts: 2671
Joined: Wed Oct 15, 2008 9:17 am

Re: Panel cointegration estimations - possible with Eviews 7

Postby EViews Glenn » Wed Jul 06, 2011 9:15 am

Pedroni's RESTAT paper will describe what you can do...

Peter Pedroni
PURCHASING POWER PARITY TESTS IN COINTEGRATED PANELS
The Review of Economics and Statistics, November 2001, 83(4): 727–731

gordeza
Posts: 10
Joined: Mon Jan 05, 2009 11:10 pm

Re: Panel cointegration estimations - possible with Eviews 7

Postby gordeza » Fri Aug 03, 2012 8:37 am

EViews Glenn wrote:Non-panel FMOLS and DOLS are supported as built-in features in EViews 7 for non-panel structured workfiles. It is possible that you could use these new FMOLS, DOLS tools to, with a bit of work, construct a panel version of the those estimators (by averaging over the individual cross-section estimates), but this is not viewed as a standard EViews 7 feature. The non-averaging forms of the estimators would be more difficult to estimate.

This is on our list of things to look at for upcoming versions.


As for the above promisse/expectation, is there any news regarding the panel FMOLS? Any help will be very much appreciated.

EViews Gareth
Fe ddaethom, fe welon, fe amcangyfrifon
Posts: 13307
Joined: Tue Sep 16, 2008 5:38 pm

Re: Panel cointegration estimations - possible with Eviews 7

Postby EViews Gareth » Fri Aug 03, 2012 8:48 am

It is still on our list for upcoming versions.
Follow us on Twitter @IHSEViews

stoddj
Posts: 48
Joined: Fri May 28, 2010 5:30 am

Re: Panel cointegration estimations - possible with Eviews 7

Postby stoddj » Sat Sep 08, 2012 3:13 pm

I understand that Eviews 7 supports Panel Cointegration testing -- although estimating the cointegrating equation itself takes a bit of additional programming, as explained in the previous posts under this heading. Eviews also supports Granger causality tests on panel data. So far, so good.

I am trying to construct a VEC model for panel data, and I have the following question: Is it possible to construct a PANEL ANALOGUE to the Granger causality test option that comes with non-panel VEC estimates? For starters, I have been trying to build up such a VEC-Granger causality program 'by hand' so that I can understand what the program is doing. Unfortunately, I can't even manage to get the same results with that!

The documentation on the Granger causality test for VEC says that only the lagged first-differenced terms are tested for exogeneity, but clearly this is a different test than the Granger causality test on a group made up of those same first-differenced terms.

I hope I've made myself reasonably clear. Any clarification will be much appreciated.

EViews Gareth
Fe ddaethom, fe welon, fe amcangyfrifon
Posts: 13307
Joined: Tue Sep 16, 2008 5:38 pm

Re: Panel cointegration estimations - possible with Eviews 7

Postby EViews Gareth » Sat Sep 08, 2012 3:37 pm

EViews doesn't support panel Granger Causality.
Follow us on Twitter @IHSEViews

stoddj
Posts: 48
Joined: Fri May 28, 2010 5:30 am

Re: Panel cointegration estimations - possible with Eviews 7

Postby stoddj » Sun Sep 09, 2012 1:18 pm

Thanks for your response, Gareth, but I must be misunderstanding something. I have here a group of two variables in a file that is clearly organized as stacked panel data -- both variables show a cross-section and date id for every observation. And under the Group View command, I find, under "Granger causality," the following output:

Pairwise Granger Causality Tests
Date: 09/09/12 Time: 16:15
Sample: 1990 2010
Lags: 3

Null Hypothesis: Obs F-Statistic Prob.

D(UNCRUP) does not Granger Cause D(PPHELGDPPC) 690 0.30091 0.8248
D(PPHELGDPPC) does not Granger Cause D(UNCRUP) 2.82598 0.0379

So are you saying that this output is not valid as a panel estimate? So that in effect, it is treating the panel as if it were one long time series?

EViews Gareth
Fe ddaethom, fe welon, fe amcangyfrifon
Posts: 13307
Joined: Tue Sep 16, 2008 5:38 pm

Re: Panel cointegration estimations - possible with Eviews 7

Postby EViews Gareth » Sun Sep 09, 2012 1:48 pm

Yes, it is doing exactly that, with the caveat that lags are handled correctly - the lag of one cross-section does not grab data from another cross-section.
Follow us on Twitter @IHSEViews

stoddj
Posts: 48
Joined: Fri May 28, 2010 5:30 am

Re: Panel cointegration estimations - possible with Eviews 7

Postby stoddj » Sun Sep 09, 2012 2:18 pm

Wow, ok -- the output sure gave me some false comfort. May I suggest that the documentation here is a little less than adequate?

Now granted, there isn't any explicit description of Granger-causality testing for panel data in the Help files -- only the Engle-Granger tests for cointegration. (But even the inclusion of that test led me to assume the Granger 'panel test' must be for real, since Granger causality and cointegration are so closely related.) It seems to me that there should be an explicit warning of some kind that these results, while available for panels, are not valid for panels as such.

But on a more constructive note, a related question: (I never expected to get so much help on a Sunday!) If one can program panel cointegration estimates by averaging the individual FMOLS from each cross-section -- is there an analogous bit of programming that one could perform via Granger causality tests on each cross-section? Thanks for your guidance.

EViews Gareth
Fe ddaethom, fe welon, fe amcangyfrifon
Posts: 13307
Joined: Tue Sep 16, 2008 5:38 pm

Re: Panel cointegration estimations - possible with Eviews 7

Postby EViews Gareth » Sun Sep 09, 2012 2:41 pm

I agree with you, although the documentation does explicitly state what calculation it is performing for Granger Causality (panel or not), and the test that is performed in a panel workfile is a valid test for causality in a panel, just, perhaps, not the test that most people would think of, although I'm not sure what the panel form of causality test that most people would think of actually is (any suggestions?).
Follow us on Twitter @IHSEViews

stoddj
Posts: 48
Joined: Fri May 28, 2010 5:30 am

Re: Panel cointegration estimations - possible with Eviews 7

Postby stoddj » Sun Sep 09, 2012 9:13 pm

It seems that there's been quite a bit of research in panel Granger tests over the last few years. One suggestion that appears very close to the existing Eviews method is found in a paper by Laszlo Konya, "Exports and growth: Granger causality analysis on OECD countries with a panel data approach," appearing in Economic Modelling 23 (2006) pp. 978–992. What this method does is to test the standard null hypothesis within each cross-section (i.e., that ALL the coefficients on all of the lags of X are zero in the equation for Y, and vice versa), but does this within a SUR context. This is to make use of any contemporaneous correlation throughout the system -- i.e., the error terms of different cross sections show contemporaneous shocks. So this seems like it would be a straightforward extension of the present E-views method.

I don't think this is a plausible assumption for the problem I'm studying at the moment, but I can imagine it works for many international trade and macro issues -- such as the paper cited.

A review of some recent approaches to this panel-Granger problem is "Testing for Granger causality in heterogeneous mixed panels," by Furkan Emirmahmutoglu and Nezir Kose, in Economic Modeling 28 (2011), pp. 870-876.

EViews Glenn
EViews Developer
Posts: 2671
Joined: Wed Oct 15, 2008 9:17 am

Re: Panel cointegration estimations - possible with Eviews 7

Postby EViews Glenn » Mon Sep 10, 2012 1:59 pm

Thanks for the references.

stoddj
Posts: 48
Joined: Fri May 28, 2010 5:30 am

Cointegration: Estimation vs. Testing

Postby stoddj » Thu Feb 07, 2013 1:13 pm

I have used FMOLS and DOLS to estimate cointegrating equations in a panel setting, via the averaging technique of Pedroni in "PURCHASING POWER PARITY TESTS IN COINTEGRATED PANELS," Review of Econ. & Stat., Nov. 2001. This approach was kindly suggested by Gareth over a year ago. Since i have the panel residuals from the FMOLS and DOLS estimates, I would hope that there would be a simple way to test these for stationarity (implying cointegration). But that doesn't appear to be the case.

Eviews 7 does have the Pedroni panel cointegration tests on residuals, but I don't see the option of residuals from DOLS or FMOLS. Looking at it from the other side, Eviews does have the eq_name.coint routine for testing cointegration on a single equation, but its not clear to me that this can be easily generalized to a panel context. Any suggestions on my next steps will be very much appreciated. (Jim Stodder)


Return to “Estimation”

Who is online

Users browsing this forum: Google [Bot] and 31 guests