panel Granger causality model

For econometric discussions not necessarily related to EViews.

Moderators: EViews Gareth, EViews Moderator

Dipti
Posts: 4
Joined: Tue Apr 10, 2012 7:53 pm

panel Granger causality model

Postby Dipti » Wed Apr 11, 2012 11:54 pm

Hi,
I'm testing panel Granger causality model for a relationship between energy use and GDP. My question is how to estimate short - run causality and long - run causality in panel data method. Please give me step how to do this in eviews.
Reply me soon
Thanks

d952
Posts: 64
Joined: Tue Nov 22, 2011 6:30 am

Re: panel Granger causality model

Postby d952 » Sat Apr 21, 2012 1:05 pm

There is not a buit in method in eviews for doing panel Granger causality test, You should apply a two step Engle Granger procedure, first perform a FMOLS or DOLS for each country, then use their residuals as ECT and apply a dynamic panel data model (panel GMM) to estimate a short and long run model!
I hope my explanation it is clear!

Dipti
Posts: 4
Joined: Tue Apr 10, 2012 7:53 pm

Re: panel Granger causality model

Postby Dipti » Mon Apr 23, 2012 4:56 am

Thank you d952.
I am a beginner of eviews and econometric, looking for your help.
Regarding you advice I did FMOLS for each country, but I am confused to use their residuals as ECT. How to find out a short - run and long - run model in dynamic panel data model (panel GMM)? Your kind help in this regard highly appreciated.

d952
Posts: 64
Joined: Tue Nov 22, 2011 6:30 am

Re: panel Granger causality model

Postby d952 » Mon Apr 23, 2012 4:49 pm

this is something that I did:

1. make a normal eviews workfile (non panel) and do FMOLS for each country and save their residuals as (ect)
2. make another eviews workfile (a panel workfile) and use dynamic panel data model (panel GMM), for instance, if you have three variables: x,y,z, then you need one more variabe (ect) so make one more series and put the name of ect on it, and copy and paste the residuals from FMOLS models (since this second workfile is panel then your series are stacked).
3. run your model by using Arrelano-Bond wizard!
4. the results that you will get, shows x,y,z,and ect, so ect is your long run result and x,y and z are your short run results.

I hope Im explaining true, this is something that I undrestood from reading different articles and did it myself.

please aks if you have more questions
Niaz

d952
Posts: 64
Joined: Tue Nov 22, 2011 6:30 am

Re: panel Granger causality model

Postby d952 » Tue Apr 24, 2012 4:00 am

By the way...if you do a google serach on `Two-step Engle-Granger procedure´ you will ger more guids about it.

Dipti
Posts: 4
Joined: Tue Apr 10, 2012 7:53 pm

Re: panel Granger causality model

Postby Dipti » Wed Apr 25, 2012 12:44 am

Thank you Niaz for your quick reply, once again I am looking for your help.

According to your advice, I did FMOLS for each country and subsequently, FMOLS result I save ECT as a residual series (proc. → make residual series, is it correct?)
For example, I have four variables: Y, F, I, P and another variable ECT (I made following your advice). I did dynamic panel data model (panel GMM).
I am confused in process so please have a look in my file (FMOLS, RESID01) one country and result (panel GMM) below.
Your kind help in this regard highly appreciated.

Thank you again.

Dependent Variable: Y
Method: Fully Modified Least Squares (FMOLS)
Sample (adjusted): 1991 2008
Included observations: 18 after adjustments
Cointegrating equation deterministics: C
Long-run covariance estimate (Prewhitening with lags = 0 from SIC
maxlags = 2, Bartlett kernel, Newey-West fixed bandwidth = 3.0000)

Variable Coefficient Std. Error t-Statistic Prob.
F -0.368394 0.510040 -0.722285 0.4820
I 0.683432 0.032544 21.00009 0.0000
P 1.687285 0.723004 2.333713 0.0350
C -17.42224 7.495042 -2.324502 0.0357


R-squared 0.977563 Mean dependent var 26.05450
Adjusted R-squared 0.972755 S.D. dependent var 0.434606
S.E. of regression 0.071736 Sum squared resid 0.072045
Durbin-Watson stat 1.513200 Long-run variance 0.002752

Year RESID01
1990 NA
1991 0.0004516896807871263
1992 0.03927098616212988
1993 0.06222345333325307
1994 0.02959863346103475
1995 0.03053206999285507
1996 -0.0006604701234316224
1997 -0.01149105627174407
1998 -0.2222864925886086
1999 -0.01227292652605172
2000 0.02380545572507842
2001 0.00272235986435021
2002 0.06177294987493554
2003 0.07943542960694216
2004 -8.379470930819366e-05
2005 -0.02936794674600307
2006 0.0160961286083996
2007 0.02153010669916355
2008 -0.01677093661993467


Result of dynamic panel data model (panel GMM)
Dependent Variable: Y
Method: Panel Generalized Method of Moments
Transformation: First Differences
Sample (adjusted): 1994 2008
Periods included: 15
Cross-sections included: 12
Total panel (balanced) observations: 180
White period instrument weighting matrix
White period standard errors & covariance (d.f. corrected)
Instrument specification: @DYN(Y,-2)
Constant added to instrument list

Variable Coefficient Std. Error t-Statistic Prob.
Y(-1) -0.030592 2.380263 -0.012852 0.9898
Y(-2) 0.316976 2.331596 0.135948 0.8920
F(-1) -2457321. 5838652. -0.420871 0.6744
F(-2) 2049434. 7141410. 0.286979 0.7745
I(-1) 3.801124 9.281843 0.409523 0.6827
I(-2) -2.780131 9.369098 -0.296734 0.7670
P(-1) 25545.08 35830.69 0.712938 0.4769
P(-2) -11064.97 28531.06 -0.387822 0.6986
ECT(-1) -2.28E+11 2.74E+11 -0.832498 0.4063
ECT(-2) -2.76E+10 1.51E+11 -0.182523 0.8554

Cross-section fixed (first differences)
Mean dependent var 1.81E+10 S.D. dependent var 5.57E+10
S.E. of regression 6.40E+10 Sum squared resid 6.97E+23
J-statistic 0.984004 Instrument rank 12

d952
Posts: 64
Joined: Tue Nov 22, 2011 6:30 am

Re: panel Granger causality model

Postby d952 » Sun Apr 29, 2012 2:43 pm

Hi, Im sorry for being late!

The process that you applied is correct initially, but in terms of doing panel GMM you should be very carefull about some issues:
1. you should be carefull about fixed effects, so its better you choose the option of time dummies (as I see in your result you didnt choose it).
2. you should be carefull about the serial correlation, Im not sure that how many methods there are for testing serial correlation in dynamic panel data method, but one is Sargan test, you can find its formula in eviews guide and do it in programming tool bar of eviews, its very easy. just search Sargan test and you will find many things about it.
3. then according to the Sargan test result you can choose the appropriate number of instruments, for instance you chosed @dyn(y,-2) but mabye its not overidentified so if the result of sargan test is not good you have to change the number of lags of independant variables and number of instrumental variables of dependant variable.

Dont hesitate to ask if you have more questions
Niaz

d952
Posts: 64
Joined: Tue Nov 22, 2011 6:30 am

Re: panel Granger causality model

Postby d952 » Sun Apr 29, 2012 3:35 pm

About FMOLS test also be carefull about the classical assumption, so consider the Durbin Watson result (in your sample is good) and also check the residualts normality... its just the same as OLS, if the probabilty is less that 0.1 then its not good!

qaiseralm
Posts: 1
Joined: Sun Jan 15, 2012 12:15 am

Re: panel Granger causality model

Postby qaiseralm » Tue May 01, 2012 12:51 am

Hi
Niyaz and everybody
thanks for your suggestion that has also resolved my issues regarding applying DOLS. i got the required result as per the used varaibles using x,y,z and ect in a panel framework. where y and z represents the short-run result and ect represents the long run result. my query is that how can i also get the long run coefficients for for Y and Z in a panel model using the same variable x,y,z and ect as mentioned earlier.
with thanks in anticipation
qaiser

Dipti
Posts: 4
Joined: Tue Apr 10, 2012 7:53 pm

Re: panel Granger causality model

Postby Dipti » Tue May 01, 2012 10:56 pm

Thank you Niaz, for your continuous help.

According to your advice, I try to do fixed effects as an option time dummies, but it shows error message (near singular matrix). I have problem on that issue, so how can I do time dummies? Please give me step. I would be grateful to your help.

Thank you again

d952
Posts: 64
Joined: Tue Nov 22, 2011 6:30 am

Re: panel Granger causality model

Postby d952 » Sun May 06, 2012 1:26 pm

Dear Qaisar
Im not sure exactly what do you mean, the ECT (error correction term)is the long run effect of y and z together on x, the procedure that I explained above is all about how to obtain ECT in a panel framework. if your meaning is that how can you obtain the long run effect of y and z on x seperately, then as I know its not possible, but if you mean something eles then let me know.

and Dipti, this is a very common problem when you apply time dummies that happen due to multicolinearity, when u add time dummies then should not add intercept (c) into your model, then let see the problem solve or not, if not Im afraid that Im not expert enough to reply your question, you can apply different options of estimation for instance rather than 2-step dynamic panel you choose n-step, but probabely it can make some other problems... I hope that some more expert people about this issue reply your question here!

d952
Posts: 64
Joined: Tue Nov 22, 2011 6:30 am

Re: panel Granger causality model

Postby d952 » Mon May 07, 2012 4:19 am

Dipti, another solution also is changing of the number of intrumental variables lags, so 1.remove intercept, 2. change the number of lags, and 3.use another estimation option (for this last one im not sure its a correct way or not) until you dont have any multicilinearity and serial correlation! hope it works

ramzan243
Posts: 8
Joined: Mon Mar 30, 2009 1:26 am

Re: panel Granger causality model

Postby ramzan243 » Sat Jul 28, 2012 7:00 am

Salam
@ d952
How ect can be the long run effect. there must be seperate coefficient for each variable.
Regards
Muhammad Ramzan

d952
Posts: 64
Joined: Tue Nov 22, 2011 6:30 am

Re: panel Granger causality model

Postby d952 » Sat Jul 28, 2012 8:45 am

Dear Mohammad

When we apply VECM model then there is seperate short run coeficients but not for the long run. ECM is error correction term of the model and indicates the long run effect of the explanatory variables on dependant variable.

Best
Niaz

ramzan243
Posts: 8
Joined: Mon Mar 30, 2009 1:26 am

Re: panel Granger causality model

Postby ramzan243 » Sat Jul 28, 2012 9:09 am

I agree , but there is also the separate long run coefficient for each variable.
Just like Engle and Granger and ARDL approach to cointegration. in these approaches there is separate long run and short run coefficients
Regards
MUHAMMAD RAMZAN


Return to “Econometric Discussions”

Who is online

Users browsing this forum: No registered users and 26 guests