2
$\begingroup$

I'm working on an analysis on GDP growth. I want to test whether regional growth in GDP in a time frame can be explained by the growth of air traffic in a preceding period (and controlling for some variables like education etc.)

Why would you use Granger causality rather than a regular regression and see if the coefficient of the variable in the preceding time frame is significant?

$\endgroup$
3
$\begingroup$

Why would you use Granger causality rather than a regular regression and see if the coefficient of the variable in the preceding time frame is significant?

You actually do that in Granger causality testing, but the "regular regression" must include the autoregressive (=own) lags per definition of Granger causality. Granger causality considers whether the lags of other variables have predictive power once the lags of the dependent variable itself are accounted for. Whether Granger causality is the right tool for a particular application is another matter; in your case, I think it is, but I am not entirely sure. It depends on whether you are interested in the predictive power of lags of some other variables on their own or in addition to the dependent variable's own lags.

$\endgroup$
  • $\begingroup$ Thanks! This clarifies it a bit already. So I'll have to add the lags of the Y variable (GDP growth) in the multiple regression? But when I add the lags of my predicting variable X (air traffic growth), why can't I just look at the significance of that coefficient? What does Granger causality add to that? $\endgroup$ – Felix Apr 1 '18 at 11:58
  • $\begingroup$ @Felix, indeed, this is what you do when you test for Granger causality (if you have already added the lags of the dependent variable). $\endgroup$ – Richard Hardy Apr 2 '18 at 19:46
  • $\begingroup$ Thanks for your time! Still one thing is confusing me: I have only two time series (t and t-1). Do I need to formulate a restricted and unrestricted model and perform a F-test? Can't I just look at the coëfficient and p-value of the unrestricted model as like in a autoregressive distributed lag model? $\endgroup$ – Felix Apr 18 '18 at 23:28
  • $\begingroup$ @Felix, Wald test can be performed on coefficients of interest in an unrestricted model. If there is only one coefficient, then it is a t-test. If there are more, then it is an F-test. A test that involves both restricted and unrestricted models would be a likelihood ratio test. $\endgroup$ – Richard Hardy Apr 18 '18 at 23:31
  • $\begingroup$ Thanks so much! So in my case (with 1 lagged coëfficient), I would be looking at the t-test, right? $\endgroup$ – Felix Apr 18 '18 at 23:39

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.