Dear Jigme, a linear model is definitely NOT sufficient to determine causality, as the association in a linear model is interpreted (when significant) as linear correlation, which is not causation. Just to give a counter-example, if both your variables are caused by a common unobserved determinant, they may end up having a strong correlation and a linear model would give you significant results. You need to test for causality separately, even though Granger causality is only one of the possible types of causality, and it assumes a relationship between the two variables that is lagged in time. If you are interested in this, you may as well consider specifying a VAR (vector autoregression) on your bivariate dataset.

I agree with the above comment. Over time growth could generate the public funds to cause more expenditure, such as in infrastructur. Or as above, an unoservable, such as technological innovation, could easily affect both. You might also consider a random effects model.

I would echo Flavio Maria Emanuele Pons comments and even go a little further on Granger causality. First, the maxim is always correlation does not equal causation. Furthermore, correlation might not even be evidence of any relationship at all. A good place to begin if you find any correlation is to conduct a cointegration test before delving further. If the series are not cointegrated then the chances are the correlation is spurious. Second, Granger causality is not a test of causality in any sense but simply a test of what comes first. If X Granger causes Y but Y does not Granger cause X, then movements in X that correlate with movements in Y occur first.

I think the answers by both Pons and John should give you an Idea. The issue of causality is not as simply as it seems. Neither correlation, test of significant impact nor Granger causality test reveals cause and effect in its real meaning. If your correlation relationship between X and Y shows positive. It only indicates that both of them are moving in the same direction but nothing on what is causing the movement. If your significant Impact test show that a unit increase in X increases Y by a certain unit, It is not really saying that without that increase in X, that Y will not exist or even change. It only says that X contributes to the change in Y. Again, If you Granger causality show that X granger cause Y, Just as john rightly answered, it only show that changes in X precedes changes in Y and not even that changes in X causes changes in Y.

To answer you question therefore, neither significant linear result nor the Granger causality can reveal the real cause-and-effect in its grammatical meaning, but since we don't have a better alternative yet, we still use what we have, taking note of the weakness in our econometric interpretation of the Granger causality and linear impact analysis result.

In your own case, the linear model alone will not do it as you may have bidirectional causality, which may lead to endogenous bisae in you result. What you should do is to first conduct the Granger causality and ensure that you have unidirectional causality from Expenditure to growth before running the linear model. Otherwise, conduct test for endogeneity and switch to appropriate models if there is endogeneity problem.

Carol Lee Graham John Ryding Flavio Maria Emanuele Pons Nathaniel E. Urama . Thank you so much for your time and valuable advice. I have considered trying it with granger causality test for now. However, i have read few papers (Pula & Elshani (2017); Medhi (2014) & Jumare, Yusuf & Mohammed (2017)and have found that they havent consider testing assumption of Autocorrelation and Heteroscedasticity. On this note, is it okay to check series stationarity and go ahead with cointegration (Johansen)and granger causality, please

And also lag selection from VAR lag selection criterion, all information criterion suggests 0 lag. How should i go ahead in testing Johansen Cointegration test with lag 0 suggested?