Granger Causality Test

[Jone]

I'm using EViews 8, running a unrestricted VAR model. Checking the impact of oil price shocks on South Africa. Using three variables government revenues, money supply and real exchange rate. When I run Granger Causality test, the result shows no causality running from oil price shocks to other variables. However, when I run Pairwise Granger Causality tests the results are completely different e.g. Oil price shocks cause government revenues and highly significant. I don't know how to solve this problem!! Which test should I use to explain the real effect.

[sakamuk]

Hi

Are you estimating an unrestricted VAR because the series are not cointegrated? , if your study found the series are not cointegrated then the estimation on the unrestricted VAR is suitable. Moreover if the series are not stationary C. W. J. Granger, Huang, and Yang (2000) recommend using differenced series

However, if the series are cointegrated then you should have estimated a VECM.


Also when you say

When I run Granger Causality test, the result shows no causality running from oil price shocks to other variables.

what kind of GC tests are you using other than the pairwise GC tests? Are you referring to the block pairwise GC tests that eviews offers?

[Jone]

Hi Sakamuk
Many thanks for your feedback. Yes, I'm running unrestricted VAR model. I've used Block GC available at EViews 8, which shows the causality between all independent variables to one dependent variable at a time. Another point is that IRF and Variance decomposition show the direction and the percentage of variation in the dependent variable, while the p value shows no significance correlation in GC test (P> 0.05). In other words, results from GC test are not in consistent with IRF and variance decomposition outputs.