can it be that two variables with out any evidence in granger causality have cointegration relationship?! i came across this problem and i don't know what is going on.
yes. It is quite possible that two variables have cointegration relationship, but not have causal relationship.
cointegration indicates long term pattern of relationship between the variables. Predictive causality indicates the predictor of one variable can be done more significantly by regressing it on lagged values of both the dependent variable and independent variable.
if two variables have causal relationship, they must have cointegration relationship. But cointegration does not ensure causality.
further discussion in this regard is most welcome.
Yes. It is quite possible. I also got such results a number of times. Additionally, I agree with sir Srikanth Potharla that " if two variables have causal relationship, they must have cointegration relationship. But cointegration does not ensure causality".
Best wishes,
Atif
I would query the previous three answers to this question. Cointegration does imply Granger Causality.
By the Granger representation theorem cointegration and VECM are equivalent. Consider the case of two I(1) variables, X and Y, which are cointegrated. Then the error correction term (ecm) must be significant in either the equation for (1) \Delta X or (2) \Delta Y or (3) in both equations. If it is significant in the equation for \Delta X then the ecm means that the lagged term in Y in the ecm is Granger causing X. Similarly if (2) then X Granger Causes Y. If (3) the causality is both ways. Of course the lagged differences may or may not contribute to Granger Causality even if the ecm is absent in the equation.
The standard test statistics for Granger Causality in the presence of I(1) do not have standard distributions. See, for example, Lutkepohl (2005), New Introduction to Multiple Time Series Analysis, Springer, pp 316-321. After reading this you should check that your software is implementing the appropriate causality tests for I(1) variables.
See also pages 71 and 91 of Engle and Granger (1991), Long-run Economic Relationships, Oxford University Press.
It's completly right. But please be informed that, two different variable can have cointegrationship, but there must be causal relationship.. You can check some articel about these anlysis
The authors listed before John C. Fran need to go back and study Engle and Granger Representation theorem of 1987. Cointegration implies Granger causality at least in one direction. Therefore, John is correct!
- Badges
- Science topic