Only train data in granger causality test?

[tim-roethig-db]

statsmodels/statsmodels/tsa/stattools.py

Line 1405 in a948e8d

def grangercausalitytests(x, maxlag, addconst=True, verbose=None):

Is the granger causality test only ran on training data? If yes, is that correct? Isn't it possible then to face issues with overfitting?

[josef-pkt]

All hypothesis tests, specification tests, and similar are only in-sample. There is no cross-validation or similar to choose the model, or number of lags.

If maxlag is large compared to the number of observations, then the test looses power, but should still be consistent.