Granger causality test and regression coefficients

[Elena Deliege]

Hello,

I am currently working on time series regressions and I was wondering which of these 2 test results should be more important to consider: the granger causality test or the regression coefficient?

Also, how could I combine the 2 to make a conclusion if they provide different results? So for instance, one variable granger causes another variable, however, their regression coefficient is not significant.

Thank you in advance.

[George Ford]

how many lags?

[Elena Deliege]

I used one lag and then I tried with 2 lags

[George Ford]

If you have 2 lags, the test is an F test, not a t-test. The individual coefficient significance levels don't tell you what you need to know. If one lag, then the two tests should give you similar results, though the test statistic is different (but related).

[Elena Deliege]

Perfect, thank you

[Jeff Wooldridge]

Elena: Are you talking about the coefficient on lags of X, or contemporaneously? I assume on the lags, as that is all that's implied by Granger Causality. And it seems George has the solution: to do a joint test of all lags of X because GC states that, once you control for enough lags of Y, no lags of X help to predict Y(t).

In the future, you should show your regression output. You get quicker, better answers that way. Please look at the FAQ.