Granger causality using panel data

[Ralph Klijn]

Dear users,

I am using Stata 15 and I have an unbalanced panel data set with time indicated by variable -year- and firms identified by -id-. Moreover, as shown below, my data set contains two variables -CSP_t_s_w- and -lag_IO_w-. Using Granger causality, I would like to test the for causality between the variables (both directions).

My dataet contains 23,097 observations and 3,077 panels.

I tried using -xtgcause-


However, the following occurs:

. set matsize 11000

. xtgcause CSP_t_s_w lag_IO_w, lags(1)
no observations
r(2000);

Does anyone know what I wrong? And anybody known how I can run a successful Granger causality test?

Thank you in advance.

* Example generated by -dataex-. To install: ssc install dataex
clear
input float year long id byte CSP_t_s_w float lag_IO_w
2000  1  1 .10273538
2001  1  2  .4753718
2002  1 11 .57919884
2003  1  8  .6090328
2004  1  9  .6398019
2005  1 12  .7047455
2006  1 12  .6007842
2007  1 12  .7196747
2008  1 12  .8020943
2009  1 12  .7530805
2010  1 11  .7809083
2011  1 10  .7956449
2012  1  4  .7915173
2005  3  0  .1929813
2006  3  1  .3637863
2007  3  1  .4739601
2008  3  1 .52113575
2009  3  1  .5178801
2010  3  0  .4741296
2011  3  0  .4554001
2012  3  0  .4432893
2003  5  0  .4057413
2004  5  2  .6909133
2005  5  2  .7740778
2006  5  2  .9303825
2009  5  2  .8186493
2010  5  0  .8578832
2003  7  1   .376464
2004  7  1  .3819983
2005  7  0   .445843
2006  7  0  .4561931
2007  7  0 .53831077
2008  7  0 .52190083
2009  7  0 .58483595
2010  7  0  .6472263
2011  7  0 .59393615
2012  7  0  .6134479
2002  8  1 .08887848
2003  8  0  .7499226
2004  8  0  .9177671
2005  8  0  .9694591
2006  8  0   .944322
2007  8  0  .8915893
2008  8  0   .935876
2009  8  0  .9408627
2010  8  1  .9994474
2011  8  1  .9193724
2012  8  0  .9837171
1991  9  5  .6455355
1992  9  6  .6606081
1993  9  6   .608819
1994  9  6  .4412391
1995  9  6 .51127815
1996  9  5  .3906025
1997  9  5   .442698
1998  9  3 .25791878
1999  9  4   .638057
2000  9  3  .6750889
2001  9  2  .4281542
2002  9  3  .5414144
2003  9  2  .6030415
2004  9  2  .6430176
2005  9  4  .7083326
2006  9  4  .7252894
2007  9  5  .7247008
2008  9  4  .6766716
2009  9  4  .6793423
2010  9  8  .7151029
2011  9  7  .6948657
2012  9  6  .6775908
2006 10  0   .276984
2007 10  0  .5215995
2008 10  0  .9288951
2009 10  0  .7196317
2010 10  0  .6809544
2007 12  0  .9596174
2009 12  0  .9458403
2011 12  0  .9803602
2009 13  1  .2162916
2010 13  0  .3018089
2004 14  1  .3291313
2005 14  0  .5344354
2006 14  1 .53080064
2007 14  1  .7165132
2008 14  1  .8001877
2009 14  1  .9331361
2010 14  0  .8769743
2011 14  0  .8963494
2012 14  2  .9074407
2006 15  4 .08887848
2007 15  4  .1131506
2008 15  4  .5344937
2009 15  4  .5535941
2010 15  0 .52943635
2002 16  0   .873021
2003 16  1  .8825788
2004 16  1   .838463
2006 16  1  .9327447
2007 16  1  .9886779
2008 16  1  .9203752
end
label values id id2
label def id2 1 "A", modify
label def id2 3 "AACC", modify
label def id2 5 "AAI", modify
label def id2 7 "AAON", modify
label def id2 8 "AAP", modify
label def id2 9 "AAPL", modify
label def id2 10 "AATI", modify
label def id2 12 "AAWW", modify
label def id2 13 "ABAT", modify
label def id2 14 "ABAX", modify
label def id2 15 "ABBC", modify
label def id2 16 "ABC", modify

[River Huang]

Hi, Ralph: I guess that -xtgcause- command requires balanced panel data. Please first (ssc) install xtbalance, and try the following:

xtset id year
xtbalance, range(2000 2012) miss( CSP_t_s_w lag_IO_w)
xtgcause CSP_t_s_w lag_IO_w, lags(1)

The result is:

Dumitrescu & Hurlin (2012) Granger non-causality test results:
--------------------------------------------------------------
Lag order: 1
W-bar =          0.3600
Z-bar =         -0.6400   (p-value = 0.5222)
Z-bar tilde =   -0.5692   (p-value = 0.5692)
--------------------------------------------------------------
H0: lag_IO_w does not Granger-cause CSP_t_s_w.
H1: lag_IO_w does Granger-cause CSP_t_s_w for at least one panelvar (id).

[Ralph Klijn]

Dear River, thank you for response,

However, I case I use (my data set runs from 1991-2012):

xtbalance, range(1991 2012) miss( CSP_t_s_w lag_IO_w)

19,929 out of 23,097 observations are deleted due to discontinues.

Moreover, the result does not show any values:

Dumitrescu & Hurlin (2012) Granger non-causality test results:
--------------------------------------------------------------
Lag order: 1
W-bar =               .
Z-bar =               .   (p-value =     .)
Z-bar tilde =         .   (p-value =     .)
--------------------------------------------------------------
H0: lag_IO_w does not Granger-cause CSP_t_s_w.
H1: lag_IO_w does Granger-cause CSP_t_s_w for at least one panelvar (id).

.

Do you happen to know a solution to this issue? Maybe another procedure?

Again, thank you for the helpful input.

[River Huang]

Hi, Ralph, Then you may want to change, say

range(2000 2012)

[Ralph Klijn]

Hello River,

Thank you for the reaction.

To be honest, I do not think it really makes a difference.

I get the following output from Stata:

. xtbalance, range(2000 2012) miss( CSP_t_s_w lag_IO_w)

(2566 observations deleted due to out of range)

(17424 observations deleted due to discontinues)

.


. xtgcause CSP_t_s_w lag_IO_w, lags(1)


Dumitrescu & Hurlin (2012) Granger non-causality test results:
--------------------------------------------------------------
Lag order: 1
W-bar =               .
Z-bar =               .   (p-value =     .)
Z-bar tilde =         .   (p-value =     .)
--------------------------------------------------------------
H0: lag_IO_w does not Granger-cause CSP_t_s_w.
H1: lag_IO_w does Granger-cause CSP_t_s_w for at least one panelvar (id).

[Ralph Klijn]

Is there any other way I can perform a Granger causality test on unbalanced panel data?

I have seen it done in papers that do research following similar methodologies and using similar data such as Dyck et al. (2017):


Reference: Dyck, A., K.V. Lins, L. Roth and H.F. Wagner (2017) ‘Do Institutional Investors Drive Corporate Social Responsibility? International Evidence’, Journal of Financial Economics, forthcoming Available at: https://papers.ssrn.com/sol3/papers....act_id=2708589

[Ralph Klijn]

Hello everyone,

Since I have not received any more response, I would like to try it once more. Hopefully someone knows the answer.

Is there a function/command in Stata that allows me to perform a Granger causality test on the unbalanced panel data set above?

Thank you all in advance.

Best regards

[Nathalie Leloup]

Hello Ralph,
I have exactly the same problem as you. When I perform xtcause, nothing appears (empty table). Please let me know if you have found any solution.
Best regards,

[Klodi Gjini]

hello River, I have more or less the same issue of an unbalanced data panel. I have two questions if you might please consider answer. 1) I want to use xtgcause for two variables ( productivity in log terms, and export in log terms ) but i get (134) problem in stata for too many values...2) What to do if when xtbalance I loose all the observations?

[Katharina Grauf]

Dear all,

I have the same problem as Ralph and Nathalie. Have you found any solution?
Please let me know.
Best regards,
Katharina

[Sam Fu]

Hello all,
I have the same problem as well with xtgcause with empty table, does anyone a solution?
Best Regards,

[Jordan Sydenham]

Dear All,

As of yet there is no simple way of running a granger causality test when the data set is unbalanced in Stata, the best you can do is to "force" the data to be balanced, however one should be careful when doing this as it may result in the P-value being wrong and your work being rejected. I would suggest that the best approach is to simply restrict the sample to only include those observations that have data available over the whole period.

Kind regards,
Jordan

[rabiya gill]

Dear all,
Can i apply the Granger causality test after panel data models? as my panel is stationary at level. So can i apply this command "xtgcause" after FE and GMM?

[rabiya gill]

because usually Granger causality test are applicable after co-integration test.


Regards
Rabia

[Toufiq Sowrov]

Dear all, my panel data set is strongly balanced. I wanted to carry out Granger causality test to find out directions for my variables. Stata shows "command xtgcause is unrecognized" when I tried the command. I need help to identify and learn my mistakes. TIA