I am comparing two time series with weekly data from 2005-2016, and wanted to see if one time series (x) predicts the other (y).

In selecting the 'order' of the Granger Test, I used the Augmented Dickey Fuller (ADF) test to find the "lag" order. The ADF was used to identify the alternative hypothesis as stationary with p < 0.05.

I then used this lag order for the Granger test order.

Am I going about this the right way?

A previous similar question is found here (below), but I thought mine was sufficiently different in that I am looking at a time series with 52 weeks/year.

Lag order for Granger causality test

  • $\begingroup$ You are not going the right way. ADF concerns unit roots and order of differencing -- but not lag order. Follow the answer of the linked thread and see the link to Dave Giles blog under the question in that thread. $\endgroup$ – Richard Hardy Sep 6 '16 at 7:20
  • $\begingroup$ I had read Dave Giles blog. To choose the order for 'm' per Dave, "both of the series are I(n) when we apply the ADF and KPSS tests, allowing for a drift and trend in each series. So, m = n." In my case: > adf.test(data_1, alternative = c("stationary", "explosive"), + k = trunc((length(data_1)-1)^(1/3))) Augmented Dickey-Fuller Test data: data_1 Dickey-Fuller = -6.4426, Lag order = 8, p-value = 0.01 alternative hypothesis: stationary $\endgroup$ – MattCrow Sep 6 '16 at 13:07
  • 2
    $\begingroup$ The lag order in the Granger causality test is the lag order of a VAR model within which the test operates. The order of integration of the series is quite a different thing. In terms of ARIMA(p,d,q) terminology, p is quite different from d. $\endgroup$ – Richard Hardy Sep 6 '16 at 13:21

Your Answer

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

Browse other questions tagged or ask your own question.