Paiwise Granger Causality Tests in VAR or VEC

[juanant]

Hello:
Somebody can help me with this test?
The User's guide explains in details what this test does when I open some series as a group (page 376) but what does in VAR or VEC?
Thanks

[EViews Gareth]

The same thing, but you don't need to provide the number of lags for the test in the VAR object, since the VAR object already has a number of lags specified.

[juanant]

Thank you for your reply.
Although I think they are diferent because each test uses diferent statistics: F when I open the data as a group and Chi-sq when I work with a VAR.

[stoddj]

Moderator Garrett says they should be the same, but I have not manage to get the same results -- that is, from the Granger test given under Lag Structure for a VAR test, and the Pairwise Granger test ust for the Group of the same two variables. I'm using the identical first-difference variables, with the same number of observations for both tests. There is also a constant term for both (as a default in the Group test). (I am not talking here about the difference between the Chi-Square and the F-test statistics, which I think should be identical in this case.) Thanks, JS

[gobbble]

I am wondering how you estimate the optimal lag length.

Fot the VEC Granger test you estimate the optimal lag length with a VAR model at level and for a VAR Grangertest you estimate the optimal lag length with a VAR model at first differences. Is that correct?

[EViews Gareth]

Moderator Garrett says they should be the same, but I have not manage to get the same results -- that is, from the Granger test given under Lag Structure for a VAR test, and the Pairwise Granger test ust for the Group of the same two variables. I'm using the identical first-difference variables, with the same number of observations for both tests. There is also a constant term for both (as a default in the Group test). (I am not talking here about the difference between the Chi-Square and the F-test statistics, which I think should be identical in this case.) Thanks, JS

Could you provide an example?

[stoddj]

Sure, here's the example. First, the two first-differenced series (don't worry, they're short):

obs D(TURN02) D(EMPCONS)
1990 NA NA
1991 NA NA
1992 NA -0.0601278679999994
1993 NA -0.0448243470000004
1994 NA 0.0201069809999996
1995 -0.0806199999999996 0.0214456070000004
1996 -0.1015800000000000 -0.0628500140000003
1997 -0.0645199999999999 -0.0668865179999995
1998 0.0552200000000002 -0.0084477390000002
1999 -0.0483700000000002 -0.0053353019999998
2000 -0.0955800000000000 0.0257837109999998
2001 -0.0683600000000002 0.0131298119999998
2002 -0.0399799999999999 -0.0119123229999997
2003 -0.0888799999999996 -0.0242382860000001
2004 0.0288500000000002 -0.0015276730000000
2005 -0.0689100000000006 0.0170763609999999
2006 -0.0320499999999999 0.0343223999999997
2007 -0.1189999999999990 0.0310736980000001
2008 0.0334599999999998 -0.0007252490000003

(Note that first 5 years on 1st series, and first 2 on the 2nd series are blank.)

Next, I now give both Granger printouts, 1st on the Group:

(1) Pairwise Granger Causality Tests
Date: 09/06/12 Time: 12:15
Sample: 1990 2008
Lags: 2

Null Hypothesis: Obs F-Statistic Prob.

D(TURN02) does not Granger Cause D(EMPCONS) 12 1.54698 0.2777
D(EMPCONS) does not Granger Cause D(TURN02) 3.01391 0.1137

... and 2nd from the VAR:

(2) VAR Granger Causality/Block Exogeneity Wald Tests
Date: 09/06/12 Time: 12:18
Sample: 1990 2008
Included observations: 12


Dependent variable: D(TURN02)

Excluded Chi-sq df Prob.

D(EMPCONS) 6.027810 2 0.0491

All 6.027810 2 0.0491


Dependent variable: D(EMPCONS)

Excluded Chi-sq df Prob.

D(TURN02) 3.093967 2 0.2129

All 3.093967 2 0.2129

[EViews Gareth]

Code: Select all

1.54698*2 = 3.093967
3.01391*2 = 6.027810


Looks the same to me.

[stoddj]

Thank you, my mistake. I guess I didn't expect the significance levels for the two tests to be so different.

[stoddj]

In regard to the earlier question about optimal lag length, i should say that i don't know the answer.

[FiliusPublii]

This thread is easy to google but I still need a bit more explanation concerning this problem, it would be great if you could help me.

I am interested in the granger causality between wage and profits and I use gdp as a control variable.

The pairwise granger causality test gives me:



So Granger causality seems to run from profits to wage, correct? That gdp is causing wage is intuitive enough. Additionally I did the VAR Granger causality test:



and I have trouble comparing the two. My VAR test statistics aren't just the pairwise test statistics times 2. Do you know why?

Thanks a lot for your answers,
Filiuspublii