So I've implemented the ADF test in a large number of variables using eviews and R (urca package, ur.df).

Althought the criteria used to select lags and the deterministic variables included were the same in both programs, multiple variables differed in the lag selection.

Most importantly, even when the lag selected and the estimated model (with trend or without etc) were the same in both programs, the test statistic estimated differed considerably (althouth in general the test result of presence of a unit root didnt change).

It made me wonder what's the difference in construction of both methods that allowed for these differences.

Does anyone know?

  • $\begingroup$ You are probably going to have to read the documentation for this. $\endgroup$ – gung - Reinstate Monica Feb 12 '16 at 19:33
  • $\begingroup$ It could be that you did not notice some subtle difference in test specifications in the different software. If the tests had identical test regressions there would be no way the answer would differ by more than a rounding error. $\endgroup$ – Richard Hardy Feb 12 '16 at 19:34

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.